Chapter 12.7

Education, ethnicity and the county shift, 2016 to 2024

In 2020 to 2024, counties with more Hispanic residents moved further toward the Republicans than other counties did (vote-weighted correlation −0.66, against +0.45 in 2012 to 2016), while the correlation with college share fell from +0.69 to +0.02.

Author: Vishal Singh
Affiliation: NYU Stern
Published: 30 September 2026

In 2012 to 2016, counties with a higher college share or a higher Hispanic share moved further toward the Democrats than other counties did. The vote-weighted correlation of a county's college share with its swing in the presidential margin was +0.69, and that of its Hispanic share was +0.45. In 2020 to 2024 the vote-weighted college correlation was +0.02 (95% CI −0.08 to 0.13), and the Hispanic correlation had reversed to −0.66, so counties with more Hispanic residents moved further toward the Republicans than other counties did. Counting each county once instead of by its votes, the college correlation stays near its 2012 to 2016 level through 2016 to 2020 and falls only in 2020 to 2024, so the pair in which the college correlation falls depends on the weighting.

A county's swing is the change in its two-party margin from one presidential election to the next, in points, and a positive swing is a move toward the Democrats. The national swing is the change in the summed margin of all counties with returns in both elections, the measure used in Chapter 10.1, Nationalization of the county vote. A correlation with the swing equals the correlation with the swing minus the national swing, so a positive value means that counties with more of a characteristic moved further toward the Democrats than other counties did, whichever way the nation moved.

A vote-weighted correlation counts each county in proportion to its two-party votes in the later election, so the largest counties count most, and a county-equal correlation counts every county once.

The characteristics are the share of adults with a bachelor's degree or more, and the Hispanic, foreign-born and non-Hispanic Black shares of residents, each measured at the start of the pair. The 95% confidence interval of an estimate follows it in parentheses, and the note beside Figure 1 says how it is computed. Chapter 12.2, The diploma map takes up the level of the vote by education, and Chapter 12.1, Five axes of the county vote ranks the axes together.The last pair, 2020 to 2024, has 3,111 matched counties, and 3,102 are used here. The 8 Connecticut counties have no county characteristics for 2020, and Jackson County, Missouri is left out because its Kansas City votes are a separate unit in 2024, which gives it a swing that is not its own. Both stay in the national swing so that it matches the other chapter.

Vote-weighted Hispanic and foreign-born correlations turned negative after 2016

Figure 1 plots the correlation of the swing with each characteristic over six pairs of elections from 2000 to 2024. The vote-weighted Hispanic correlation is positive in 2004 to 2008, 2008 to 2012 and 2012 to 2016 (+0.45 in the last) and negative in 2016 to 2020 (−0.48 (95% CI −0.62 to −0.20)) and 2020 to 2024 (−0.66 (95% CI −0.76 to −0.46)). The foreign-born correlation follows the same path, from +0.58 in 2012 to 2016 to −0.50 and −0.69. The college correlation has a positive point estimate in all six pairs in both weightings and is highest, vote-weighted, in 2012 to 2016, though the intervals for 2020 to 2024 (vote-weighted) and 2008 to 2012 (county-equal, +0.09 (95% CI −0.01 to 0.20)) include zero.The intervals resample whole states, because counties in one state share candidates, campaigns and election law, and a resample of counties would give a narrower interval. The resampling repeats the calculation on many draws of the states with replacement, which is a bootstrap. The returns are full counts, so an interval says how much a correlation depends on which states are in the set and is not sampling error of the vote. The characteristics are census and American Community Survey estimates interpolated to the election years, so they carry sampling error that no interval includes.

Figure 1

Vote-weighted, the college correlation fell after 2016 while the Hispanic and foreign-born correlations turned negative

Correlation between a county characteristic at the start of a pair of presidential elections and the county's swing in the two-party margin, 6 pairs from 2000 to 2004 through 2020 to 2024

Correlation of the characteristic with county swing

Show data

Data for Vote-weighted, the college correlation fell after 2016 while the Hispanic and foreign-born correlations turned negative
Pair of elections	Characteristic	Weighting	Correlation	Interval low	Interval high	Counties
2000–04	College share	Vote-weighted	+0.44	+0.31	+0.57	3,113
2004–08	College share	Vote-weighted	+0.25	+0.11	+0.38	3,113
2008–12	College share	Vote-weighted	+0.10	+0.02	+0.18	3,113
2012–16	College share	Vote-weighted	+0.69	+0.62	+0.78	3,112
2016–20	College share	Vote-weighted	+0.25	+0.11	+0.41	3,113
2020–24	College share	Vote-weighted	+0.02	−0.08	+0.13	3,102
2000–04	Hispanic share	Vote-weighted	−0.03	−0.15	+0.09	3,113
2004–08	Hispanic share	Vote-weighted	+0.27	+0.07	+0.40	3,113
2008–12	Hispanic share	Vote-weighted	+0.33	+0.17	+0.46	3,113
2012–16	Hispanic share	Vote-weighted	+0.45	+0.31	+0.53	3,112
2016–20	Hispanic share	Vote-weighted	−0.48	−0.62	−0.20	3,113
2020–24	Hispanic share	Vote-weighted	−0.66	−0.76	−0.46	3,102
2000–04	Foreign-born share	Vote-weighted	+0.01	−0.18	+0.24	3,113
2004–08	Foreign-born share	Vote-weighted	+0.21	+0.05	+0.35	3,113
2008–12	Foreign-born share	Vote-weighted	+0.45	+0.28	+0.56	3,113
2012–16	Foreign-born share	Vote-weighted	+0.58	+0.44	+0.69	3,112
2016–20	Foreign-born share	Vote-weighted	−0.50	−0.67	−0.10	3,113
2020–24	Foreign-born share	Vote-weighted	−0.69	−0.77	−0.49	3,102
2000–04	Black share	Vote-weighted	+0.05	−0.06	+0.15	3,113
2004–08	Black share	Vote-weighted	+0.03	−0.09	+0.16	3,113
2008–12	Black share	Vote-weighted	+0.43	+0.34	+0.53	3,113
2012–16	Black share	Vote-weighted	+0.16	+0.03	+0.31	3,112
2016–20	Black share	Vote-weighted	−0.13	−0.29	0.00	3,113
2020–24	Black share	Vote-weighted	−0.07	−0.25	+0.09	3,102
2000–04	College share	County-equal	+0.40	+0.30	+0.49	3,113
2004–08	College share	County-equal	+0.34	+0.26	+0.41	3,113
2008–12	College share	County-equal	+0.09	−0.01	+0.20	3,113
2012–16	College share	County-equal	+0.57	+0.49	+0.66	3,112
2016–20	College share	County-equal	+0.51	+0.45	+0.60	3,113
2020–24	College share	County-equal	+0.16	+0.07	+0.24	3,102
2000–04	Hispanic share	County-equal	−0.09	−0.21	+0.07	3,113
2004–08	Hispanic share	County-equal	+0.19	+0.11	+0.29	3,113
2008–12	Hispanic share	County-equal	+0.20	+0.11	+0.30	3,113
2012–16	Hispanic share	County-equal	+0.33	+0.25	+0.42	3,112
2016–20	Hispanic share	County-equal	−0.30	−0.51	+0.10	3,113
2020–24	Hispanic share	County-equal	−0.39	−0.51	−0.16	3,102
2000–04	Foreign-born share	County-equal	+0.05	−0.05	+0.16	3,113
2004–08	Foreign-born share	County-equal	+0.24	+0.15	+0.33	3,113
2008–12	Foreign-born share	County-equal	+0.26	+0.17	+0.35	3,113
2012–16	Foreign-born share	County-equal	+0.51	+0.44	+0.58	3,112
2016–20	Foreign-born share	County-equal	−0.03	−0.20	+0.22	3,113
2020–24	Foreign-born share	County-equal	−0.34	−0.47	−0.14	3,102
2000–04	Black share	County-equal	−0.08	−0.21	+0.04	3,113
2004–08	Black share	County-equal	−0.04	−0.17	+0.09	3,113
2008–12	Black share	County-equal	+0.49	+0.36	+0.59	3,113
2012–16	Black share	County-equal	+0.29	+0.20	+0.38	3,112
2016–20	Black share	County-equal	−0.12	−0.26	−0.02	3,113
2020–24	Black share	County-equal	−0.24	−0.36	−0.14	3,102

The vote-weighted Hispanic and foreign-born lines cross zero between 2012 to 2016 and 2016 to 2020, from +0.45 to −0.48 and from +0.58 to −0.50, while the college line stays above zero and falls. Points above zero mean that counties with more of the characteristic swung further toward the Democrats than other counties did in that pair, and points below zero mean they swung further toward the Republicans. Each point is a Pearson correlation across matched counties, weighted by two-party votes in the later election in the first view and counting each county once in the second. Bands are 95% intervals from resampling states, drawn for college and Hispanic share only.Note: Election returns are full counts of votes, so they carry no sampling error. The 95% intervals come from a bootstrap that resamples whole states, so they show how much a number depends on which states are in the set. They do not include the sampling error of the county characteristics, which are census and American Community Survey estimates interpolated between years.Note: The pair 2020 to 2024 leaves out the eight Connecticut counties, which have no county characteristics for 2020, and Jackson County, Missouri, whose swing is not its own.Source: County presidential returns, 2000-2024 (Algara and Amlani; MIT Election Data and Science Lab), full counts of votes, and county demographics (decennial census and American Community Survey estimates interpolated to election years); weighted by two-party votes in the later election or county-equal, as stated

The spread of county swings was also smaller in 2020 to 2024. The vote-weighted standard deviation of county swings, measured as in Chapter 10.1, Nationalization of the county vote, was 10.8 points in 2012 to 2016 and 4.5 in 2020 to 2024, so the correlations describe how the counties are ordered and not how far they moved. The differences in points are in Figure 4, where the 103 counties with a Hispanic share of 50% or more averaged −7.6 (95% CI −12.2 to −5.3) points relative to the national swing, vote-weighted.Counting each county once, the Hispanic correlation also changes sign, from +0.33 in 2012 to 2016 to −0.30 (95% CI −0.51 to 0.10) in 2016 to 2020, though the later interval includes zero, and the foreign-born correlation is −0.03 (95% CI −0.20 to 0.22) in 2016 to 2020 and −0.34 (95% CI −0.47 to −0.14) in 2020 to 2024. Black share has a vote-weighted correlation of +0.16 (95% CI 0.03 to 0.31) in 2012 to 2016 and −0.07 (95% CI −0.25 to 0.09) in 2020 to 2024, and counting each county once it is −0.24 (95% CI −0.36 to −0.14) in 2020 to 2024.

One account of the 2016 campaign holds that the parties' coalitions were already dividing along racial and ethnic identities and that the Trump campaign stressed race, immigration and religion [1]. A study that combined survey and county data found the educational composition of counties to be among the strongest county-level indicators of shifts in the presidential vote from 2000 to 2020 [2]. The correlations here do not test either account.

Counties furthest below the national swing had high Hispanic shares

Figure 2 colors each county by its swing minus the national swing, for 2012 to 2016, 2016 to 2020 and 2020 to 2024. The national swing in 2020 to 2024 was −6.1 points, a move toward the Republicans, and 75.6% of counties were within 5 points of it, against 24.6% in 2012 to 2016. Among the 584 counties with 50,000 or more votes in 2024, the ten furthest below the national swing all have a Hispanic share of 28.0% or more and a foreign-born share of 22.9% or more. They lie along the Rio Grande (Webb, Hidalgo, Cameron and El Paso counties), in the Imperial Valley, in the New York City area (Bronx, Queens, Passaic and Hudson counties) and in Miami-Dade. Without the vote cut, the ten furthest below still all have a Hispanic share of 28.0% or more, and four small Texas counties (Maverick, Starr, Dimmit and Culberson) enter the ten, which brings the lowest foreign-born share among them down to 9.0%. Of the ten furthest above the national swing, 5 are in Georgia, and the swing of Henry County was +15.2 points relative to the nation. This chapter does not explain why the Georgia counties sit at that end.

Figure 2

In 2020 to 2024, 75.6% of counties were within 5 points of the national swing, and the counties furthest below it were high-Hispanic

County swing in the Democratic minus Republican margin, minus the national swing, in percentage points, for the pairs 2012 to 2016, 2016 to 2020 and 2020 to 2024

Loading county values

County swing minus national swing (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.

Find a county

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

Alaska draws as no data. Its election returns are reported by election district, and the district codes cannot be matched to boroughs and census areas.

Show data

Data for In 2020 to 2024, 75.6% of counties were within 5 points of the national swing, and the counties furthest below it were high-Hispanic
Pair of elections	Position	County	Swing minus national swing (points)	Swing (points)	Hispanic share (%)	Foreign-born share (%)	College share (%)	Two-party votes, later election
2012–16	Ten furthest below the national swing	Schuylkill, PA	−29.3	−31.2	3.2	1.8	14.6	60,771
2012–16	Ten furthest below the national swing	Trumbull, OH	−28.0	−29.9	1.5	1.5	17.2	92,038
2012–16	Ten furthest below the national swing	St. Clair, MI	−24.1	−26.1	3.0	2.7	16.6	73,604
2012–16	Ten furthest below the national swing	Mahoning, OH	−23.4	−25.4	5.0	3.0	21.9	110,997
2012–16	Ten furthest below the national swing	Luzerne, PA	−23.0	−25.0	8.1	5.2	21.3	130,395
2012–16	Ten furthest below the national swing	Monroe, MI	−22.4	−24.4	3.2	2.1	18.2	70,124
2012–16	Ten furthest below the national swing	Jefferson, MO	−22.3	−24.2	1.7	1.7	17.7	100,519
2012–16	Ten furthest below the national swing	Lackawanna, PA	−22.1	−24.0	5.6	4.9	25.6	100,367
2012–16	Ten furthest below the national swing	Androscoggin, ME	−21.4	−23.3	1.7	3.0	19.6	51,164
2012–16	Ten furthest below the national swing	Fayette, PA	−21.3	−23.2	0.9	0.9	14.4	52,536
2012–16	Ten furthest above the national swing	Salt Lake, UT	+35.0	+33.0	17.4	12.3	31.8	313,906
2012–16	Ten furthest above the national swing	Davis, UT	+28.3	+26.4	8.7	4.8	34.7	90,995
2012–16	Ten furthest above the national swing	Utah, UT	+25.7	+23.8	10.9	7.1	36.9	130,704
2012–16	Ten furthest above the national swing	Arlington, VA	+25.5	+23.6	15.2	23.0	72.5	112,202
2012–16	Ten furthest above the national swing	Weber, UT	+21.4	+19.4	17.2	7.3	22.7	63,366
2012–16	Ten furthest above the national swing	Alexandria, VA	+20.2	+18.2	16.4	26.4	61.3	70,527
2012–16	Ten furthest above the national swing	Fairfax, VA	+19.6	+17.7	15.8	29.8	59.2	512,843
2012–16	Ten furthest above the national swing	Travis, TX	+18.7	+16.7	33.6	17.7	45.3	435,469
2012–16	Ten furthest above the national swing	Orange, CA	+17.5	+15.5	33.9	30.4	37.5	1,106,504
2012–16	Ten furthest above the national swing	Gwinnett, GA	+17.4	+15.5	20.3	25.1	34.7	313,142
2016–20	Ten furthest below the national swing	Webb, TX	−32.2	−29.6	95.5	26.0	18.1	67,718
2016–20	Ten furthest below the national swing	Hidalgo, TX	−27.2	−24.7	91.9	27.1	18.4	218,726
2016–20	Ten furthest below the national swing	Miami-Dade, FL	−25.4	−22.9	67.3	53.1	29.0	1,150,697
2016–20	Ten furthest below the national swing	Cameron, TX	−22.9	−20.4	89.4	23.3	17.5	113,095
2016–20	Ten furthest below the national swing	Imperial, CA	−21.8	−19.2	83.7	31.0	14.7	55,525
2016–20	Ten furthest below the national swing	Bronx, NY	−15.0	−12.5	55.7	34.8	19.7	423,114
2016–20	Ten furthest below the national swing	Osceola, FL	−14.5	−12.0	52.5	20.0	20.9	170,777
2016–20	Ten furthest below the national swing	El Paso, TX	−12.2	−9.7	82.2	24.6	22.9	262,457
2016–20	Ten furthest below the national swing	Queens, NY	−11.7	−9.1	27.9	47.2	32.0	781,703
2016–20	Ten furthest below the national swing	Kings, NY	−10.6	−8.1	19.1	36.4	36.5	906,082
2016–20	Ten furthest above the national swing	Henry, GA	+13.6	+16.2	6.7	7.4	27.8	121,702
2016–20	Ten furthest above the national swing	Forsyth, GA	+13.5	+16.0	9.4	16.5	51.6	127,331
2016–20	Ten furthest above the national swing	Douglas, GA	+11.8	+14.3	9.4	8.7	27.1	68,268
2016–20	Ten furthest above the national swing	Kaufman, TX	+11.6	+14.1	20.8	7.7	19.8	56,029
2016–20	Ten furthest above the national swing	Hamilton, IN	+11.3	+13.9	3.9	8.1	57.8	189,977
2016–20	Ten furthest above the national swing	El Paso, CO	+11.0	+13.6	16.8	7.0	37.6	364,769
2016–20	Ten furthest above the national swing	Fayette, GA	+10.7	+13.3	7.2	9.7	46.3	71,018
2016–20	Ten furthest above the national swing	Collin, TX	+10.7	+13.3	15.2	20.2	51.7	483,263
2016–20	Ten furthest above the national swing	Paulding, GA	+10.5	+13.0	6.1	5.8	24.3	84,212
2016–20	Ten furthest above the national swing	Stafford, VA	+10.5	+13.0	12.3	9.8	39.3	77,881
2020–24	Ten furthest below the national swing	Imperial, CA	−19.7	−25.8	85.3	30.9	15.7	52,629
2020–24	Ten furthest below the national swing	Webb, TX	−19.6	−25.7	95.3	25.4	20.1	65,336
2020–24	Ten furthest below the national swing	Bronx, NY	−16.4	−22.5	55.6	34.6	21.3	359,844
2020–24	Ten furthest below the national swing	Queens, NY	−14.9	−21.0	28.0	47.2	34.8	701,910
2020–24	Ten furthest below the national swing	El Paso, TX	−14.3	−20.4	82.7	23.5	25.0	248,280
2020–24	Ten furthest below the national swing	Hidalgo, TX	−14.0	−20.1	92.1	26.2	20.0	215,277
2020–24	Ten furthest below the national swing	Passaic, NJ	−13.6	−19.7	43.1	32.6	30.2	196,110
2020–24	Ten furthest below the national swing	Cameron, TX	−13.0	−19.1	89.6	22.9	20.1	115,249
2020–24	Ten furthest below the national swing	Miami-Dade, FL	−12.8	−18.9	68.7	54.3	32.5	1,085,945
2020–24	Ten furthest below the national swing	Hudson, NJ	−11.9	−18.0	41.7	43.2	46.4	224,678
2020–24	Ten furthest above the national swing	Henry, GA	+15.2	+9.2	7.7	8.3	28.8	128,235
2020–24	Ten furthest above the national swing	Douglas, GA	+12.3	+6.2	11.1	9.3	29.5	70,236
2020–24	Ten furthest above the national swing	Kaufman, TX	+12.3	+6.2	24.9	9.5	23.3	68,789
2020–24	Ten furthest above the national swing	Paulding, GA	+11.3	+5.2	7.7	7.1	26.7	94,571
2020–24	Ten furthest above the national swing	Island, WA	+10.7	+4.6	8.4	7.7	35.4	50,562
2020–24	Ten furthest above the national swing	Henderson, NC	+10.4	+4.3	11.6	7.5	34.0	68,858
2020–24	Ten furthest above the national swing	Newton, GA	+10.3	+4.2	6.5	6.8	22.9	58,732
2020–24	Ten furthest above the national swing	Mesa, CO	+9.9	+3.8	15.1	4.1	29.9	89,412
2020–24	Ten furthest above the national swing	Fayette, GA	+9.8	+3.7	8.0	10.6	48.3	73,999
2020–24	Ten furthest above the national swing	Buncombe, NC	+9.8	+3.7	7.6	6.4	43.6	157,678

Red counties moved further toward the Republicans than the nation did (the national swing was −6.1 points in 2020 to 2024) and blue counties less far, so blue does not mean a move toward the Democrats. The average county's swing minus the national swing is +2.7 points, counting each county once, and 75.6% of counties fall in the palest class, within 5 points of the national swing, against 24.6% in 2012 to 2016. Among counties with 50,000 or more votes, the reddest in 2020 to 2024 lie along the Rio Grande, in the Imperial Valley, in the New York City area and in Miami-Dade.Note: Election returns are full counts of votes, so the map draws no intervals. Counties are drawn on 2020 boundaries and matched across elections on FIPS code; a county with no return in one of the two elections draws as no data. Jackson County, Missouri is left blank in 2020 to 2024 because Kansas City's votes are a separate unit in 2024. The data table lists the ten counties at each end for each pair, not the whole map.Source: County presidential returns, 2012-2024 (Algara and Amlani; MIT Election Data and Science Lab), full counts of votes, not weighted; tooltip characteristics from the county demographics panel (decennial census and American Community Survey estimates interpolated to election years)

The 2012 to 2016 frame has different places at its ends. Among the counties with 50,000 or more votes, 8 of the ten furthest below the national swing are in Pennsylvania, Ohio or Michigan, and 7 of the ten furthest above are in Utah or northern Virginia.

The college fall depends on the weighting

Vote-weighted, the college correlation has already fallen in 2016 to 2020 (+0.25 (95% CI 0.11 to 0.41)). Counting each county once, it is +0.57 in 2012 to 2016 and +0.51 in 2016 to 2020 and falls only in 2020 to 2024, to +0.16 (95% CI 0.07 to 0.24), an interval that excludes zero, so only the vote-weighted series reaches zero. The third view of Figure 1 shows the two series together. The county-equal correlation is not a fixed fraction of the vote-weighted one. The two differ by more than 0.2 in 4 of the 24 comparisons of a characteristic and a pair, by as much as 0.47, and they have opposite signs in 2 cells where both are near zero.

A vote-weighted correlation counts the largest counties most: in 2020 to 2024 the 134 counties with 250,000 or more votes cast 46.3% of the votes. Figure 3 shows that, inside each size group and counting each county once, the 2020 to 2024 college correlation is similar, +0.23 among the large counties and +0.22 among the rest. The groups differ in level. The large counties have a higher mean college share (40.7%, against 22.8%) and a lower mean swing (−5.5 points, against −3.3), and the vote-weighted correlation across all counties, +0.02, is lower than either group's.

Figure 3

The 2020 to 2024 college correlation is similar in large and other counties and is near zero only across all counties weighted by votes

Correlation of the college share and the foreign-born share with the county swing, with the mean college share and mean swing of each group of counties, for 2016 to 2020 and 2020 to 2024

Counties and weighting	Pair of elections	College correlation	95% interval	Foreign-born correlation	95% interval	Mean college share, start of pair (%)	Mean swing (points)
All counties, vote-weighted	2016–20	+0.25	+0.11 to +0.41	−0.50	−0.67 to −0.10	32.2	+2.5
All counties, county-equal	2016–20	+0.51	+0.45 to +0.60	−0.03	−0.20 to +0.22	21.7	+1.1
Large counties, county-equal	2016–20	+0.21	+0.02 to +0.38	−0.65	−0.78 to −0.37	37.6	+3.2
Other counties, county-equal	2016–20	+0.53	+0.48 to +0.63	−0.03	−0.24 to +0.26	20.9	+1.0
All counties, vote-weighted	2020–24	+0.02	−0.08 to +0.13	−0.69	−0.77 to −0.49	34.6	−4.8
All counties, county-equal	2020–24	+0.16	+0.07 to +0.24	−0.34	−0.47 to −0.14	23.6	−3.4
Large counties, county-equal	2020–24	+0.23	0.00 to +0.40	−0.70	−0.80 to −0.54	40.7	−5.5
Other counties, county-equal	2020–24	+0.22	+0.15 to +0.29	−0.28	−0.40 to −0.07	22.8	−3.3

Show data

Data for The 2020 to 2024 college correlation is similar in large and other counties and is near zero only across all counties weighted by votes
Counties and weighting	Pair of elections	Counties	Share of two-party votes (%)	College correlation	College, low	College, high	Foreign-born correlation	Foreign-born, low	Foreign-born, high	Mean college share, start of pair (%)	Mean swing (points)
All counties, vote-weighted	2016–20	3,113	100.0	+0.25	+0.11	+0.41	−0.50	−0.67	−0.10	32.2	+2.5
All counties, county-equal	2016–20	3,113	100.0	+0.51	+0.45	+0.60	−0.03	−0.20	+0.22	21.7	+1.1
Large counties, county-equal	2016–20	142	48.8	+0.21	+0.02	+0.38	−0.65	−0.78	−0.37	37.6	+3.2
Other counties, county-equal	2016–20	2,971	51.2	+0.53	+0.48	+0.63	−0.03	−0.24	+0.26	20.9	+1.0
All counties, vote-weighted	2020–24	3,102	100.0	+0.02	−0.08	+0.13	−0.69	−0.77	−0.49	34.6	−4.8
All counties, county-equal	2020–24	3,102	100.0	+0.16	+0.07	+0.24	−0.34	−0.47	−0.14	23.6	−3.4
Large counties, county-equal	2020–24	134	46.3	+0.23	0.00	+0.40	−0.70	−0.80	−0.54	40.7	−5.5
Other counties, county-equal	2020–24	2,968	53.7	+0.22	+0.15	+0.29	−0.28	−0.40	−0.07	22.8	−3.3

Read down the college column for 2020 to 2024: the two size groups have similar correlations, and the vote-weighted correlation across all counties is lower because the large counties have a higher college share and a lower swing. Large counties are those with 250,000 or more two-party votes in the later election. Inside a size group each county counts once; the all-county rows give both weightings.Note: Intervals resample whole states; see Methods. Intervals within a size group resample states from the counties of that group.Note: Counties are those in the correlations for the pair; the eight Connecticut counties and, in 2020 to 2024, Jackson County, Missouri are left out.Source: County presidential returns, 2000-2024 (Algara and Amlani; MIT Election Data and Science Lab), full counts of votes, and county demographics (decennial census and American Community Survey estimates interpolated to election years); weighted by two-party votes in the later election or county-equal, as stated

A statement that the college correlation is near zero in 2020 to 2024 therefore holds for the vote-weighted correlation across all counties and not for the correlation inside either size group.

The Hispanic gradient steepens at high shares

Figure 4 plots each county's 2020 to 2024 swing minus the national swing against its Hispanic share, with the mean of four groups of counties. The vote-weighted mean is +3.3 (95% CI 2.8 to 3.9) points among the 2,285 counties under 10% Hispanic, +1.7 (95% CI 0.3 to 3.4) among the 503 counties from 10% to under 25%, −1.9 (95% CI −3.6 to −0.3) among the 211 counties from 25% to under 50% and −7.6 (95% CI −12.2 to −5.3) among the 103 counties at 50% or more. Those last counties moved −13.7 (95% CI −18.3 to −11.4) points in their own margin, against −6.1 nationally. Counting each county once, the four means are +3.1 (95% CI 2.6 to 3.5), +2.9 (95% CI 2.1 to 3.7), +1.0 (95% CI −0.7 to 1.9) and −3.9 (95% CI −7.5 to −2.9), and only the last has an interval below zero. In both weightings the drop from one group to the next gets larger as the Hispanic share rises. A single correlation (−0.66 vote-weighted, −0.39 county-equal) summarizes a relationship that is shallow where Hispanic shares are low and steeper where they are high.

Figure 4

County swing relative to the nation fell with Hispanic share in 2020 to 2024, and fell faster at higher shares

Each dot is a county with characteristics for 2020; the vertical axis is its swing in the Democratic minus Republican margin between 2020 and 2024, minus the national swing, in percentage points

County swing minus national swing, 2020 to 2024 (points)

Dot area: Two-party votes, 2024360,0001,300,0003,600,000

Show data

Data for County swing relative to the nation fell with Hispanic share in 2020 to 2024, and fell faster at higher shares
County	Hispanic share, 2020 (%)	College share, 2020 (%)	Swing minus national swing (points)	Two-party votes, 2024
Autauga, AL	3.3	28.7	+4.5	27,923
Baldwin, AL	5.2	32.8	+1.9	120,732
Barbour, AL	5.3	11.1	−1.0	9,764
Bibb, AL	3.1	11.5	−0.4	9,191
Blount, AL	9.8	14.5	+5.2	27,930
Bullock, AL	7.0	9.2	+2.1	4,084
Butler, AL	1.6	14.7	−0.9	8,423
Calhoun, AL	4.2	19.8	+0.5	48,106
Chambers, AL	3.3	14.7	−1.5	14,116
Cherokee, AL	1.8	14.7	+3.5	12,911
Chilton, AL	9.0	15.0	+1.5	19,618
Choctaw, AL	0.9	14.6	−2.2	6,618
Clarke, AL	0.9	14.4	+1.0	11,892
Clay, AL	3.2	13.2	−1.3	6,727
Cleburne, AL	2.3	16.5	+3.2	7,593
Coffee, AL	8.8	22.1	+1.5	22,096
Colbert, AL	3.2	19.5	−1.3	26,851
Conecuh, AL	0.8	13.6	−0.5	6,003
Coosa, AL	1.9	12.7	−3.6	5,236
Covington, AL	1.9	16.4	+1.9	16,991
Crenshaw, AL	2.2	18.6	−0.6	6,457
Cullman, AL	4.7	17.6	+3.3	42,743
Dale, AL	6.9	19.3	+0.3	18,960
Dallas, AL	0.6	17.1	+1.0	15,426
DeKalb, AL	15.9	14.0	+2.2	29,391
Elmore, AL	3.2	25.6	+2.4	41,148
Escambia, AL	2.4	13.0	−2.8	14,848
Etowah, AL	4.7	17.9	+0.9	45,680
Fayette, AL	0.7	13.6	+1.5	8,300
Franklin, AL	19.1	15.4	−1.2	11,985
Geneva, AL	4.7	13.2	+3.0	12,320
Greene, AL	1.4	12.6	−1.2	4,018
Hale, AL	0.7	16.6	−6.0	7,237
Henry, AL	2.4	20.4	−1.6	9,252
Houston, AL	4.0	22.6	+1.2	43,821
Jackson, AL	3.3	16.1	+2.4	23,349
Jefferson, AL	4.7	35.1	+3.3	293,235
Lamar, AL	1.8	11.7	+2.3	6,839
Lauderdale, AL	3.1	27.4	−0.5	43,034
Lawrence, AL	2.7	15.5	−1.5	16,007
Lee, AL	4.6	39.4	−1.6	71,818
Limestone, AL	6.7	27.6	+5.1	52,468
Lowndes, AL	0.3	14.5	−2.5	5,625
Macon, AL	1.3	20.8	−1.6	7,766
Madison, AL	5.9	45.1	+5.1	193,254
Marengo, AL	2.6	18.3	+1.0	9,626
Marion, AL	2.9	13.3	+2.5	13,442
Marshall, AL	15.7	21.1	+3.5	39,987
Mobile, AL	3.2	25.2	+1.5	172,814
Monroe, AL	0.3	12.6	−1.2	9,747
Montgomery, AL	4.2	34.1	+5.3	88,423
Morgan, AL	9.5	22.3	+3.0	52,841
Perry, AL	1.1	16.8	+0.5	4,443
Pickens, AL	4.2	13.8	−1.0	8,853
Pike, AL	2.5	28.8	−1.7	13,123
Randolph, AL	2.9	17.7	0.0	11,022
Russell, AL	5.8	17.8	+1.3	20,500
St. Clair, AL	2.7	19.7	+6.2	43,141
Shelby, AL	6.6	45.5	+5.9	112,753
Sumter, AL	0.9	18.7	−1.3	5,267
Talladega, AL	2.4	15.9	−2.1	32,998
Tallapoosa, AL	2.7	19.6	−0.1	19,859
Tuscaloosa, AL	4.7	31.6	+0.5	84,123
Walker, AL	3.2	15.0	+2.5	29,566
Washington, AL	1.0	15.7	−0.7	8,397
Wilcox, AL	0.2	12.3	0.0	5,242
Winston, AL	3.4	13.8	+4.6	11,075
Apache, AZ	6.5	14.5	−8.8	31,667
Cochise, AZ	35.1	26.8	+2.6	58,232
Coconino, AZ	14.9	39.0	+1.7	69,080
Gila, AZ	18.5	19.0	+2.7	27,405
Graham, AZ	31.9	14.7	+2.9	15,044
Greenlee, AZ	47.0	18.8	−0.8	3,262
La Paz, AZ	26.6	12.6	+0.9	7,571
Maricopa, AZ	31.2	35.0	+0.4	2,031,547
Mohave, AZ	16.9	14.3	+1.9	109,764
Navajo, AZ	11.3	17.6	−2.9	50,234
Pima, AZ	37.0	35.2	+2.5	507,119
Pinal, AZ	30.1	21.7	+1.4	207,582
Santa Cruz, AZ	82.9	22.0	−11.0	18,964
Yavapai, AZ	15.0	28.3	+1.6	148,063
Yuma, AZ	64.5	16.1	−8.2	67,568
Arkansas, AR	3.7	15.0	+1.8	5,454
Ashley, AR	6.0	12.5	+2.0	6,919
Baxter, AR	2.7	18.6	+3.0	20,594
Benton, AR	17.6	36.2	+5.7	125,138
Boone, AR	2.9	17.5	+3.4	16,822
Bradley, AR	15.7	14.6	−1.8	3,174
Calhoun, AR	3.6	15.3	−2.3	2,053
Carroll, AR	15.4	20.6	+3.4	11,282
Chicot, AR	6.1	16.3	−2.6	3,454
Clark, AR	5.2	25.7	−0.2	7,485
Clay, AR	1.7	14.5	+5.2	4,875
Cleburne, AR	2.7	17.6	+4.8	12,536
Cleveland, AR	0.7	17.8	+0.6	3,328
Columbia, AR	3.1	21.2	+1.4	7,833
Conway, AR	4.4	18.8	+1.9	8,342
Craighead, AR	5.7	28.4	+4.1	36,362
Crawford, AR	8.3	19.1	+4.7	23,368
Crittenden, AR	3.0	17.2	+1.0	14,390
Cross, AR	2.2	16.5	+4.7	6,395
Dallas, AR	1.4	11.2	+0.2	2,280
Desha, AR	6.8	13.3	−1.2	3,443
Drew, AR	4.0	24.1	0.0	6,253
Faulkner, AR	4.9	33.5	+3.4	53,109
Franklin, AR	3.3	12.9	+5.0	6,814
Fulton, AR	1.3	14.6	+1.3	4,946
Garland, AR	6.6	24.1	+3.9	41,374
Grant, AR	2.9	20.9	+4.6	7,947
Greene, AR	3.3	17.1	+5.0	15,552
Hempstead, AR	14.1	16.2	+0.9	5,969
Hot Spring, AR	3.8	16.4	+2.7	12,044
Howard, AR	12.6	15.0	+1.7	4,404
Independence, AR	7.7	18.3	+5.4	13,712
Izard, AR	2.7	17.1	+2.7	5,803
Jackson, AR	3.1	13.2	+1.5	4,692
Jefferson, AR	2.4	19.0	+4.2	21,270
Johnson, AR	13.9	17.2	+4.1	8,873
Lafayette, AR	1.8	11.5	+2.5	2,287
Lawrence, AR	1.9	14.5	+2.5	5,573
Lee, AR	2.9	8.0	+4.7	2,451
Lincoln, AR	4.6	10.2	+0.3	3,315
Little River, AR	3.8	15.7	+1.4	4,828
Logan, AR	3.3	14.5	+3.9	8,031
Lonoke, AR	5.0	21.3	+6.1	30,015
Madison, AR	6.2	12.5	+3.2	7,376
Marion, AR	2.5	15.5	+3.1	7,731
Miller, AR	3.8	16.5	+1.9	15,611
Mississippi, AR	4.6	14.6	−3.0	10,537
Monroe, AR	2.0	12.7	+4.8	2,387
Montgomery, AR	4.6	15.9	+2.9	3,632
Nevada, AR	2.4	12.9	−1.4	2,851
Newton, AR	1.3	15.8	+4.5	3,707
Ouachita, AR	2.5	14.9	+0.7	8,468
Perry, AR	3.3	17.1	+2.2	4,482
Phillips, AR	2.0	15.2	−0.3	4,852
Pike, AR	7.3	18.7	+1.2	4,306
Poinsett, AR	3.5	11.4	+2.8	6,966
Polk, AR	6.8	16.5	+4.2	8,132
Pope, AR	10.1	24.6	+4.2	23,605
Prairie, AR	2.1	16.7	+1.3	3,152
Pulaski, AR	7.3	36.6	+5.7	150,015
Randolph, AR	2.4	12.9	+4.1	6,505
St Francis, AR	4.5	12.6	+1.6	5,862
Saline, AR	6.1	28.5	+7.3	56,345
Scott, AR	7.9	12.6	+3.5	3,338
Searcy, AR	2.4	13.0	+3.1	3,816
Sebastian, AR	15.2	23.3	+4.2	44,371
Sevier, AR	34.5	11.4	−1.3	4,634
Sharp, AR	2.4	12.2	+4.1	7,294
Stone, AR	2.1	14.2	+2.3	5,897
Union, AR	4.4	20.1	+2.5	15,215
Van Buren, AR	3.3	15.4	+2.8	7,460
Washington, AR	17.8	34.2	+3.3	94,022
White, AR	4.7	20.6	+3.9	30,155
Woodruff, AR	0.8	15.6	+1.6	2,273
Yell, AR	20.9	14.1	+4.8	6,360
Alameda, CA	22.8	50.5	−1.7	640,340
Alpine, CA	14.6	39.7	+6.6	722
Amador, CA	15.2	20.3	+2.5	21,801
Butte, CA	18.4	30.1	+1.2	91,407
Calaveras, CA	13.5	21.8	+1.6	25,806
Colusa, CA	61.3	15.6	−5.9	6,845
Contra Costa, CA	26.7	44.3	−0.9	511,316
Del Norte, CA	20.0	18.4	+5.2	10,265
El Dorado, CA	13.7	37.6	+2.8	108,812
Fresno, CA	53.9	23.2	−6.4	317,552
Glenn, CA	43.6	14.8	−2.0	10,164
Humboldt, CA	13.0	32.0	+1.3	61,359
Imperial, CA	85.3	15.7	−19.7	52,629
Inyo, CA	23.6	28.9	+2.9	8,669
Kern, CA	55.2	17.9	−5.1	276,120
Kings, CA	56.6	14.3	−4.9	40,593
Lake, CA	23.0	17.3	−1.8	25,955
Lassen, CA	20.9	12.9	+3.2	11,097
Los Angeles, CA	48.4	34.8	−5.0	3,606,971
Madera, CA	59.6	16.6	−3.4	53,325
Marin, CA	17.8	60.6	+4.0	140,206
Mariposa, CA	13.2	29.4	+3.0	9,247
Mendocino, CA	26.6	24.6	−2.8	37,577
Merced, CA	61.8	14.7	−9.2	84,145
Modoc, CA	15.3	21.5	+3.9	3,892
Mono, CA	27.1	32.1	+4.2	5,816
Monterey, CA	60.3	27.2	−5.3	142,286
Napa, CA	35.2	38.5	+0.7	63,569
Nevada, CA	10.1	39.6	+3.6	59,961
Orange, CA	34.0	42.7	−0.4	1,346,546
Placer, CA	15.0	42.0	+4.1	227,899
Plumas, CA	9.8	23.5	+5.7	9,745
Riverside, CA	50.3	24.4	−3.3	915,459
Sacramento, CA	23.8	32.8	+0.6	633,704
San Benito, CA	61.2	22.1	−5.9	26,881
San Bernardino, CA	54.5	22.3	−7.0	740,530
San Diego, CA	34.3	41.3	+0.1	1,434,642
San Francisco, CA	15.7	59.6	−0.3	386,313
San Joaquin, CA	42.1	21.0	−9.0	255,643
San Luis Obispo, CA	23.9	38.0	+3.9	146,246
San Mateo, CA	24.5	52.9	−0.7	319,573
Santa Barbara, CA	47.0	35.9	+0.8	179,019
Santa Clara, CA	25.0	55.0	−0.8	721,668
Santa Cruz, CA	34.4	42.9	+0.9	128,976
Shasta, CA	11.1	22.9	+2.6	86,669
Sierra, CA	10.5	22.6	+3.2	1,707
Siskiyou, CA	13.2	22.7	+2.4	20,790
Solano, CA	28.2	27.9	−1.2	184,342
Sonoma, CA	28.5	37.5	+1.1	243,026
Stanislaus, CA	48.5	18.7	−6.0	192,333
Sutter, CA	32.1	20.5	−9.3	38,388
Tehama, CA	27.3	18.4	−0.2	25,918
Trinity, CA	6.8	20.8	+1.9	5,428
Tulare, CA	65.9	15.6	−7.1	135,075
Tuolumne, CA	13.1	23.5	+2.9	28,119
Ventura, CA	43.6	35.0	+0.1	376,325
Yolo, CA	32.5	43.7	+1.2	89,249
Yuba, CA	29.8	17.9	+1.8	29,216
Adams, CO	41.6	27.1	−1.4	227,080
Alamosa, CO	47.6	28.5	−4.3	7,301
Arapahoe, CO	20.6	45.0	+1.5	316,036
Archuleta, CO	17.5	40.3	+7.9	9,122
Baca, CO	11.1	23.3	+5.1	1,961
Bent, CO	32.4	15.4	+0.8	2,141
Boulder, CO	14.3	63.6	+5.6	190,907
Broomfield, CO	13.4	58.2	+7.3	45,497
Chaffee, CO	9.9	41.3	+13.0	14,026
Cheyenne, CO	10.6	24.8	+3.6	1,038
Clear Creek, CO	7.5	54.2	+9.8	5,916
Conejos, CO	51.0	25.6	−4.5	3,985
Costilla, CO	58.0	24.4	−6.5	2,005
Crowley, CO	30.5	10.6	+6.0	1,653
Custer, CO	7.9	38.9	+7.1	3,771
Delta, CO	14.8	24.2	+7.6	18,978
Denver, CO	28.8	53.4	+1.0	353,399
Dolores, CO	12.0	27.1	+4.8	1,401
Douglas, CO	9.6	60.3	+6.3	237,859
Eagle, CO	29.8	50.7	+0.6	27,091
Elbert, CO	8.3	35.1	+5.3	19,977
El Paso, CO	18.1	40.3	+7.2	370,530
Fremont, CO	13.3	20.2	+7.5	24,839
Garfield, CO	30.6	32.8	+5.9	29,621
Gilpin, CO	8.0	48.6	+9.7	3,983
Grand, CO	9.7	41.4	+7.1	9,565
Gunnison, CO	9.8	61.1	+5.1	10,576
Hinsdale, CO	5.7	45.7	+9.5	589
Huerfano, CO	32.2	25.0	+0.2	4,311
Jackson, CO	10.8	31.3	+8.1	807
Jefferson, CO	15.8	49.2	+7.0	351,380
Kiowa, CO	8.7	20.6	+8.3	846
Kit Carson, CO	19.5	19.0	+1.9	3,639
Lake, CO	29.4	44.9	+2.0	3,756
La Plata, CO	12.9	46.7	+7.1	34,701
Larimer, CO	12.3	50.3	+8.3	219,056
Las Animas, CO	39.9	19.6	+2.4	7,604
Lincoln, CO	18.6	19.4	+4.3	2,522
Logan, CO	17.1	20.4	+5.2	9,953
Mesa, CO	15.1	29.9	+9.9	89,412
Mineral, CO	5.1	52.9	+7.3	734
Moffat, CO	16.4	20.4	+6.9	6,525
Montezuma, CO	12.6	30.2	+7.5	14,765
Montrose, CO	21.1	26.6	+9.9	25,058
Morgan, CO	37.3	17.5	−1.3	13,121
Otero, CO	42.1	19.4	+1.9	8,684
Ouray, CO	4.4	52.8	+7.8	4,015
Park, CO	7.0	37.1	+6.6	11,669
Phillips, CO	29.7	28.5	+2.2	2,300
Pitkin, CO	10.8	62.0	−1.4	10,924
Prowers, CO	39.5	18.9	+1.2	5,060
Pueblo, CO	42.7	24.3	−0.9	83,016
Rio Blanco, CO	10.6	24.7	+10.3	3,625
Rio Grande, CO	42.5	30.2	+1.3	6,049
Routt, CO	8.0	51.8	+6.4	15,884
Saguache, CO	36.4	26.9	−4.3	3,281
San Juan, CO	21.1	42.8	+13.7	564
San Miguel, CO	11.3	60.7	+1.7	4,683
Sedgwick, CO	19.2	23.9	+6.3	1,331
Summit, CO	15.8	54.3	+3.7	17,006
Teller, CO	7.4	36.7	+5.8	15,921
Washington, CO	11.1	22.9	+5.0	2,765
Weld, CO	30.3	30.4	+3.1	175,221
Yuma, CO	27.1	23.0	+7.2	4,565
Kent, DE	7.7	26.0	+4.0	86,680
New Castle, DE	11.0	38.2	+1.5	271,568
Sussex, DE	10.4	31.2	+6.4	145,861
District Of Columbia, DC	11.5	62.0	+3.8	315,261
Alachua, FL	11.4	46.4	−0.1	134,517
Baker, FL	3.4	16.1	+3.5	14,908
Bay, FL	7.9	27.1	+2.4	96,698
Bradford, FL	4.7	12.7	+1.8	13,866
Brevard, FL	11.4	32.6	+1.7	357,766
Broward, FL	31.5	34.9	−6.7	866,280
Calhoun, FL	5.6	11.3	+0.8	6,388
Charlotte, FL	7.9	25.4	−1.3	122,930
Citrus, FL	6.4	19.8	+1.3	97,632
Clay, FL	10.8	27.2	+4.1	125,637
Collier, FL	28.1	38.1	−2.3	214,987
Columbia, FL	7.2	15.9	+1.1	33,358
DeSoto, FL	30.8	11.5	−4.9	12,413
Dixie, FL	4.4	10.1	+1.7	8,103
Duval, FL	11.3	32.6	+0.7	465,650
Escambia, FL	6.4	28.4	+1.7	161,008
Flagler, FL	11.1	28.5	−1.4	79,445
Franklin, FL	5.6	24.2	−0.5	6,701
Gadsden, FL	11.4	19.8	+0.2	21,698
Gilchrist, FL	7.0	13.8	+2.0	10,593
Glades, FL	23.4	12.5	−1.0	5,256
Gulf, FL	4.7	23.4	+2.6	8,654
Hamilton, FL	10.6	9.6	−1.2	5,691
Hardee, FL	43.1	10.6	−5.2	8,087
Hendry, FL	56.0	10.2	−9.3	13,349
Hernando, FL	15.5	20.1	−0.8	109,877
Highlands, FL	21.3	19.0	−0.3	51,609
Hillsborough, FL	29.7	36.3	−4.0	663,472
Holmes, FL	3.3	11.2	+5.0	9,075
Indian River, FL	13.2	32.2	+0.3	98,391
Jackson, FL	5.2	14.0	−1.1	21,966
Jefferson, FL	4.6	21.2	−5.7	8,440
Lafayette, FL	14.5	9.6	+1.7	3,737
Lake, FL	17.4	26.1	+1.5	225,046
Lee, FL	23.3	29.8	−3.2	389,901
Leon, FL	7.4	47.8	−0.5	154,917
Levy, FL	9.5	15.3	+1.5	24,239
Liberty, FL	5.5	15.0	−0.4	3,464
Madison, FL	5.9	13.9	−3.3	9,105
Manatee, FL	17.4	32.8	−1.4	227,160
Marion, FL	15.4	22.0	+0.4	212,609
Martin, FL	14.8	35.3	−0.5	97,660
Miami-Dade, FL	68.7	32.5	−12.8	1,085,945
Monroe, FL	24.6	36.9	−4.4	43,997
Nassau, FL	5.1	33.0	+5.2	65,088
Okaloosa, FL	10.4	33.5	+3.1	112,383
Okeechobee, FL	26.0	15.1	−3.3	15,986
Orange, FL	33.0	37.4	−3.5	599,086
Osceola, FL	55.3	26.5	−9.3	170,918
Palm Beach, FL	23.5	38.8	−6.0	739,348
Pasco, FL	17.2	26.9	+0.9	315,229
Pinellas, FL	10.6	34.6	+0.6	511,924
Polk, FL	26.4	22.0	−0.2	345,923
Putnam, FL	10.7	14.0	−0.4	36,054
St. Johns, FL	8.2	47.7	+1.4	195,621
St. Lucie, FL	20.6	24.6	−1.5	183,810
Santa Rosa, FL	6.3	29.6	+2.1	111,349
Sarasota, FL	10.0	38.6	−1.7	275,887
Seminole, FL	22.9	41.9	−0.3	250,452
Sumter, FL	6.1	33.7	+4.6	104,685
Suwannee, FL	10.7	16.2	+1.9	21,778
Taylor, FL	4.3	13.7	+0.4	9,945
Union, FL	5.6	10.8	+3.4	6,195
Volusia, FL	15.4	26.4	−1.6	307,823
Wakulla, FL	4.4	20.9	+2.7	19,687
Walton, FL	7.5	32.5	+0.1	49,257
Washington, FL	3.9	11.8	+1.9	12,510
Appling, GA	10.4	10.4	+0.9	8,321
Atkinson, GA	26.7	10.6	−0.8	3,050
Bacon, GA	8.4	9.8	+5.9	4,831
Baker, GA	4.4	19.0	+2.0	1,473
Baldwin, GA	2.6	24.3	+2.6	18,733
Banks, GA	7.2	17.2	+6.4	10,494
Barrow, GA	13.4	20.7	+9.3	43,679
Bartow, GA	9.8	22.0	+6.2	57,213
Ben Hill, GA	6.5	12.0	+0.4	6,480
Berrien, GA	5.7	14.5	+3.1	8,050
Bibb, GA	4.1	25.8	+4.6	68,830
Bleckley, GA	3.5	18.9	+4.0	6,024
Brantley, GA	2.7	10.7	+5.3	8,480
Brooks, GA	6.4	16.1	+0.1	7,189
Bryan, GA	7.9	34.5	+5.3	24,517
Bulloch, GA	4.7	30.0	+1.0	32,499
Burke, GA	3.4	15.0	−1.5	11,021
Butts, GA	3.6	12.7	+4.7	12,968
Calhoun, GA	5.2	12.4	+2.9	2,053
Camden, GA	7.3	23.5	+1.6	26,224
Candler, GA	12.3	13.7	+0.9	4,562
Carroll, GA	7.8	22.9	+4.3	60,170
Catoosa, GA	3.5	23.6	+7.1	34,854
Charlton, GA	11.3	12.6	+0.9	4,614
Chatham, GA	7.4	36.0	+5.2	140,094
Chattahoochee, GA	17.1	28.3	+3.3	1,685
Chattooga, GA	5.5	10.8	+4.3	10,665
Cherokee, GA	11.5	39.0	+6.7	160,980
Clarke, GA	11.3	46.4	+2.0	52,346
Clay, GA	0.2	20.2	+2.8	1,434
Clayton, GA	13.8	20.8	+4.1	111,080
Clinch, GA	6.3	9.9	+2.2	2,903
Cobb, GA	14.1	49.6	+6.6	397,083
Coffee, GA	12.6	13.0	+1.1	15,683
Colquitt, GA	20.1	14.6	+3.3	16,565
Columbia, GA	7.4	37.5	+6.5	85,281
Cook, GA	6.7	16.8	+0.3	7,330
Coweta, GA	7.6	34.7	+8.0	85,315
Crawford, GA	3.7	12.6	+2.7	6,324
Crisp, GA	3.6	14.5	+5.2	8,092
Dade, GA	2.6	17.1	+4.6	8,147
Dawson, GA	6.1	35.0	+9.2	19,465
Decatur, GA	6.6	14.6	−0.8	11,512
DeKalb, GA	9.5	46.5	+3.4	362,252
Dodge, GA	3.7	16.2	+1.9	8,330
Dooly, GA	7.7	13.9	+4.5	4,164
Dougherty, GA	3.1	22.5	+7.0	33,735
Douglas, GA	11.1	29.5	+12.3	70,236
Early, GA	1.5	14.2	−0.3	4,876
Echols, GA	24.6	11.8	+0.3	1,434
Effingham, GA	5.7	25.5	+7.1	36,087
Elbert, GA	5.9	14.0	−0.7	9,560
Emanuel, GA	4.7	12.4	+0.7	9,592
Evans, GA	12.1	15.3	+0.7	4,225
Fannin, GA	2.9	23.1	+6.2	16,039
Fayette, GA	8.0	48.3	+9.8	73,999
Floyd, GA	11.9	23.6	+5.5	44,493
Forsyth, GA	9.8	56.0	+6.3	136,790
Franklin, GA	5.2	17.4	+3.4	12,106
Fulton, GA	7.7	56.5	+4.5	529,407
Gilmer, GA	12.0	23.1	+7.4	18,389
Glascock, GA	2.4	13.7	+2.1	1,667
Glynn, GA	7.2	31.1	+3.4	43,702
Gordon, GA	16.2	17.4	+5.5	27,477
Grady, GA	12.4	15.9	−0.2	10,675
Greene, GA	6.6	34.2	+3.7	12,729
Gwinnett, GA	22.5	38.2	+4.3	415,548
Habersham, GA	15.8	22.9	+5.7	23,178
Hall, GA	28.8	25.6	+5.9	101,338
Hancock, GA	1.1	9.8	−2.5	4,228
Haralson, GA	2.0	17.1	+6.1	16,304
Harris, GA	4.4	35.2	+4.6	22,259
Hart, GA	4.0	19.0	+1.0	14,274
Heard, GA	2.8	11.0	+3.0	6,194
Henry, GA	7.7	28.8	+15.2	128,235
Houston, GA	7.1	31.7	+7.4	80,997
Irwin, GA	5.7	16.2	+3.0	4,326
Jackson, GA	9.5	25.6	+9.5	46,969
Jasper, GA	4.5	17.1	+1.1	9,084
Jeff Davis, GA	13.2	12.1	+1.7	5,859
Jefferson, GA	3.7	12.1	−2.0	7,439
Jenkins, GA	7.8	7.7	+1.6	3,396
Johnson, GA	2.2	7.7	−0.3	3,979
Jones, GA	2.0	20.4	+2.1	16,038
Lamar, GA	3.0	16.6	+1.5	10,370
Lanier, GA	6.9	19.5	+1.8	3,721
Laurens, GA	3.0	18.1	+1.7	23,280
Lee, GA	3.2	30.6	+7.4	17,612
Liberty, GA	12.7	19.9	−0.8	22,900
Lincoln, GA	1.7	14.7	−1.1	4,910
Long, GA	12.2	14.0	+3.3	7,033
Lowndes, GA	6.5	27.1	+0.2	47,568
Lumpkin, GA	5.4	31.5	+3.1	17,695
McDuffie, GA	3.4	16.4	+0.5	10,499
McIntosh, GA	1.5	18.7	−1.5	7,375
Macon, GA	4.5	10.5	+0.9	4,671
Madison, GA	6.6	20.1	+4.7	16,704
Marion, GA	7.9	15.6	+2.5	3,601
Meriwether, GA	1.6	12.6	+1.2	11,748
Miller, GA	1.9	20.0	+2.3	2,715
Mitchell, GA	4.9	13.0	+0.3	8,851
Monroe, GA	2.6	26.8	+2.5	17,643
Montgomery, GA	7.2	18.2	+3.2	3,960
Morgan, GA	3.5	26.9	+2.1	13,122
Murray, GA	15.6	12.8	+4.1	17,424
Muscogee, GA	8.1	29.0	+5.3	80,029
Newton, GA	6.5	22.9	+10.3	58,732
Oconee, GA	5.6	53.2	+3.9	27,044
Oglethorpe, GA	5.8	22.5	+2.7	8,770
Paulding, GA	7.7	26.7	+11.3	94,571
Peach, GA	8.6	22.4	+4.7	13,397
Pickens, GA	3.6	24.7	+6.6	20,803
Pierce, GA	5.1	16.8	+4.0	9,744
Pike, GA	2.0	20.5	+4.1	12,512
Polk, GA	13.6	14.1	+2.9	19,101
Pulaski, GA	4.3	13.8	+4.6	4,317
Putnam, GA	6.7	27.3	+5.0	12,832
Quitman, GA	1.1	11.9	+0.3	1,136
Rabun, GA	8.4	29.8	+7.0	10,373
Randolph, GA	3.8	14.3	+4.6	2,974
Richmond, GA	5.4	23.5	+4.8	83,129
Rockdale, GA	10.5	29.8	+12.7	44,876
Schley, GA	3.9	17.1	+2.6	2,423
Screven, GA	2.5	14.7	−0.1	6,906
Seminole, GA	2.2	14.4	+0.7	4,002
Spalding, GA	5.4	18.3	+10.3	32,863
Stephens, GA	3.8	16.6	+2.4	13,036
Stewart, GA	10.0	13.2	+3.2	2,024
Sumter, GA	6.0	20.7	+3.5	12,005
Talbot, GA	0.7	12.3	−2.5	3,371
Taliaferro, GA	4.4	8.9	−0.8	882
Tattnall, GA	11.3	14.7	+1.7	8,482
Taylor, GA	0.9	15.0	+2.1	3,966
Telfair, GA	13.2	8.5	−2.3	4,204
Terrell, GA	2.8	11.6	+1.7	4,328
Thomas, GA	3.8	23.4	+1.6	22,017
Tift, GA	12.6	19.1	+4.3	16,934
Toombs, GA	11.8	17.5	+0.9	10,882
Towns, GA	3.1	31.8	+4.5	8,804
Treutlen, GA	1.7	14.1	−0.8	3,114
Troup, GA	4.2	21.9	+3.7	31,149
Turner, GA	4.8	10.3	+2.5	3,822
Twiggs, GA	1.3	10.4	−1.2	4,444
Union, GA	3.6	24.8	+7.1	17,786
Upson, GA	2.5	14.5	+0.6	13,626
Walker, GA	2.6	20.2	+6.6	31,898
Walton, GA	5.6	23.5	+9.7	58,012
Ware, GA	4.6	14.8	+3.6	14,347
Warren, GA	0.6	11.4	−0.7	2,586
Washington, GA	2.4	16.0	+3.4	9,467

Showing the first 500 of 3102 rows.

Dots fall to the right as the Hispanic share rises, and the drop from one group to the next gets larger. The solid line joins the vote-weighted means of four groups of counties (under 10% Hispanic, 10% to under 25%, 25% to under 50%, and 50% or more), each drawn at its group's median share, and the dashed line does the same counting each county once; whiskers are 95% intervals from resampling states and dot area is two-party votes in 2024. The correlation with the swing is −0.66 vote-weighted and −0.39 county-equal.Note: Intervals resample whole states; see Methods. The number of counties in each group is in the tooltip.Note: Counties without 2020 characteristics (the eight Connecticut counties) and Jackson County, Missouri are left out. Points are shown for all remaining counties.Source: County presidential returns, 2000-2024 (Algara and Amlani; MIT Election Data and Science Lab), full counts of votes, and county demographics (decennial census and American Community Survey estimates interpolated to election years); weighted by two-party votes in the later election or county-equal, as stated

The college-share view of the same figure shows little slope in the vote-weighted means. The vote-weighted mean is +1.4 (95% CI 0.2 to 2.5) points among counties under 20% college, +1.1 (95% CI −0.2 to 2.8) from 20% to under 40% and +1.7 (95% CI 0.4 to 3.3) at 40% or more. Counting each county once, it is +1.9 (95% CI 1.4 to 2.5), +3.3 (95% CI 2.7 to 3.8) and +3.2 (95% CI 2.1 to 4.3), so the county-equal correlation of +0.16 is a difference between the counties with the lowest college shares and the rest.

The four characteristics are correlated across counties, so Figure 5 compares each correlation with its standardized coefficient in one regression on all four. Coefficient intervals use standard errors clustered by state. In 2020 to 2024, vote-weighted, college share goes from +0.02 alone to +0.31 (95% CI 0.24 to 0.38) with the others fixed. Among counties with the same Hispanic, foreign-born and Black shares, a county with a college share one standard deviation higher had a swing +0.31 standard deviations further toward the Democrats. The foreign-born coefficient is −0.69 (95% CI −0.93 to −0.44) and the Hispanic coefficient −0.18 (95% CI −0.35 to −0.01). Vote-weighted, the Hispanic coefficient is smaller as characteristics are added: −0.66 alone, −0.35 (95% CI −0.50 to −0.20) in a regression on Hispanic and foreign-born share only, and −0.18 with all four. The R-squared of a regression is the share of the variance of county swings that its fitted values reproduce. Vote-weighted, it is 60.2% for the four characteristics together in 2020 to 2024 and 0.0% for college share alone, and 28.1% and 2.4% counting each county once. In 2012 to 2016 the vote-weighted values were 70.6% and 47.3%, and the county-equal values 58.3% and 32.2%.

Figure 5

With the other three shares held fixed, college share tracks the 2020 to 2024 swing although its vote-weighted correlation alone is near zero

Correlation of each characteristic with the county swing, alone and as a standardized coefficient in one regression on the four characteristics; two pairs of elections and two weightings

Correlation, characteristic alone
Coefficient with all four characteristics

Show data

Data for With the other three shares held fixed, college share tracks the 2020 to 2024 swing although its vote-weighted correlation alone is near zero
Characteristic	Pair	Weighting	Correlation alone	Alone, low	Alone, high	Coefficient, all four	Together, low	Together, high	Counties
College share	2016–20	County-equal	+0.51	+0.45	+0.60	+0.51	+0.43	+0.59	3,113
Hispanic share	2016–20	County-equal	−0.30	−0.51	+0.10	−0.30	−0.60	+0.01	3,113
Foreign-born share	2016–20	County-equal	−0.03	−0.20	+0.22	−0.02	−0.19	+0.15	3,113
Black share	2016–20	County-equal	−0.12	−0.26	−0.02	−0.10	−0.17	−0.04	3,113
College share	2020–24	County-equal	+0.16	+0.07	+0.24	+0.24	+0.15	+0.34	3,102
Hispanic share	2020–24	County-equal	−0.39	−0.51	−0.16	−0.22	−0.36	−0.08	3,102
Foreign-born share	2020–24	County-equal	−0.34	−0.47	−0.14	−0.29	−0.48	−0.10	3,102
Black share	2020–24	County-equal	−0.24	−0.36	−0.14	−0.24	−0.35	−0.13	3,102
College share	2016–20	Vote-weighted	+0.25	+0.11	+0.41	+0.56	+0.46	+0.65	3,113
Hispanic share	2016–20	Vote-weighted	−0.48	−0.62	−0.20	+0.08	−0.16	+0.31	3,113
Foreign-born share	2016–20	Vote-weighted	−0.50	−0.67	−0.10	−0.77	−0.99	−0.55	3,113
Black share	2016–20	Vote-weighted	−0.13	−0.29	0.00	−0.09	−0.16	−0.01	3,113
College share	2020–24	Vote-weighted	+0.02	−0.08	+0.13	+0.31	+0.24	+0.38	3,102
Hispanic share	2020–24	Vote-weighted	−0.66	−0.76	−0.46	−0.18	−0.35	−0.01	3,102
Foreign-born share	2020–24	Vote-weighted	−0.69	−0.77	−0.49	−0.69	−0.93	−0.44	3,102
Black share	2020–24	Vote-weighted	−0.07	−0.25	+0.09	−0.04	−0.16	+0.07	3,102

In 2020 to 2024, vote-weighted, college share goes from +0.02 alone to +0.31 with the other three held fixed, and Hispanic share from −0.66 to −0.18; foreign-born share is −0.69 alone and −0.69 with the others held fixed. The open dot is the correlation of a characteristic with the county swing by itself, and the filled dot is its standardized coefficient when all four are in one regression; the last column is the filled dot minus the open dot. Intervals are 95% intervals: bootstrap over states for correlations, errors clustered by state for coefficients.Note: Intervals resample whole states; see Methods.Note: Each characteristic is standardized with the weighted mean and standard deviation of the weighting shown, so a coefficient from a regression on one characteristic equals its correlation. The four characteristics are correlated across counties, so a coefficient depends on which others are in the regression. Non-Hispanic white share is not in the regression.Source: County presidential returns, 2000-2024 (Algara and Amlani; MIT Election Data and Science Lab), full counts of votes, and county demographics (decennial census and American Community Survey estimates interpolated to election years); weighted by two-party votes in the later election or county-equal, as stated

In 2016 to 2020, vote-weighted, college share also goes from +0.25 alone to +0.56 (95% CI 0.46 to 0.65), the foreign-born coefficient is −0.77 (95% CI −0.99 to −0.55) and the Hispanic coefficient is +0.08 (95% CI −0.16 to 0.31), an interval that includes zero. The Hispanic correlation in that pair, −0.48, goes to −0.24 (95% CI −0.42 to −0.07) with foreign-born share added and to +0.08 with all four. Counting each county once in 2016 to 2020, the coefficients are close to the correlations, +0.51 for college share and −0.30 for Hispanic share. Counting each county once in 2020 to 2024, the coefficients are +0.24 for college share, −0.22 for Hispanic share, −0.29 for foreign-born share and −0.24 for Black share. Black share has a coefficient different from zero in that version but not in the vote-weighted one (−0.04 (95% CI −0.16 to 0.07)).

ZIP code areas reproduce the Hispanic and foreign-born correlations

The 2016 to 2020 swing can be measured for 29,323 ZIP code areas, which are far smaller than counties, using precinct votes allocated to each area [3]. The two college measures are not identical. The county panel reports a bachelor's degree share, which averages 21.7% across counties, and the ZIP code atlas reports a college-education share with no documented definition that averages 54.7% across areas, so the college correlations are not like for like. Figure 6 compares the vote-weighted correlations at the two grains. The Hispanic correlation is −0.50 (95% CI −0.60 to −0.29) in ZIP code areas and −0.48 (95% CI −0.62 to −0.20) in counties, and the foreign-born correlation is −0.46 (95% CI −0.61 to −0.14) and −0.50 (95% CI −0.67 to −0.10). The Hispanic and foreign-born correlations are reproduced at the finer grain. The college correlation is +0.47 (95% CI 0.42 to 0.55) in ZIP code areas, against +0.25 (95% CI 0.11 to 0.41) in counties. The 14 areas with a missing characteristic are dropped.

Figure 6

ZIP code areas give Hispanic and foreign-born correlations similar to counties for 2016 to 2020, and a larger college correlation with a different college measure

Vote-weighted correlation of four characteristics with the swing in the Democratic minus Republican margin between 2016 and 2020, at county and ZIP code area grain

Counties
ZIP code areas

Show data

Data for ZIP code areas give Hispanic and foreign-born correlations similar to counties for 2016 to 2020, and a larger college correlation with a different college measure
Characteristic	County	County, low	County, high	Counties	ZIP code area	ZIP, low	ZIP, high	ZIP code areas
College share	+0.25	+0.11	+0.41	3,113	+0.47	+0.42	+0.55	29,323
Hispanic share	−0.48	−0.62	−0.20	3,113	−0.50	−0.60	−0.29	29,323
Foreign-born share	−0.50	−0.67	−0.10	3,113	−0.46	−0.61	−0.14	29,323
Black share	−0.13	−0.29	0.00	3,113	−0.18	−0.29	−0.11	29,323

Dots that sit close together mean the county correlation is reproduced at the finer grain, as for Hispanic and foreign-born share; the college dots differ in level and in measure, because counties use the bachelor's degree share and ZIP code areas use the atlas's college-education share.Note: Intervals resample whole states; see Methods.Note: ZIP code area votes are estimates allocated from precinct returns, so they carry allocation error. There is no ZIP code area release for 2024, so the 2020 to 2024 pair cannot be checked below the county. The 14 areas for which the atlas codes one of the four characteristics as missing, with a negative value, are dropped.Source: County presidential returns, 2000-2024 (Algara and Amlani; MIT Election Data and Science Lab), full counts of votes, and county demographics (decennial census and American Community Survey estimates interpolated to election years); weighted by two-party votes in the later election or county-equal, as stated; ZIP code areas from precinct votes allocated to census geographies (Fekrazad 2025), 2016 and 2020, weighted by 2020 presidential votes

No ZIP code area returns exist for 2024, so the pair in which the vote-weighted college correlation is near zero, 2020 to 2024, cannot be checked below the county.

High-Hispanic counties swung furthest toward the Republicans

Figure 7 sorts the 2020 counties by rules, applied in this order, into four groups. High-Hispanic counties are 25% Hispanic or more. Of the rest, high-Black counties are 25% non-Hispanic Black or more, and affluent suburban counties have 35% or more of adults holding a bachelor's degree, a median household income of $75,000 or more and fewer than 25% of residents in rural areas. Every other county is in a fourth group. The figure plots each group's mean swing minus the national swing, minus the mean over all counties in the same weighting. That all-county mean is +1.3 (95% CI 0.2 to 2.8) points vote-weighted and +2.7 (95% CI 2.2 to 3.2) county-equal. It is not zero because the national swing is the change in the summed margin, which large counties dominate, so each group is compared with the all-county mean and not with zero.

Figure 7

High-Hispanic counties swung 4.2 points further toward the Republicans than the average county, vote-weighted

Mean county swing in the Democratic minus Republican margin minus the national swing, minus the mean over all counties of the same weighting, in percentage points, by county type, weighted by votes and counting each county once

Vote-weighted
County-equal

Mean county swing minus the all-county mean of the same weighting, 2020 to 2024 (points)

Show data

Data for High-Hispanic counties swung 4.2 points further toward the Republicans than the average county, vote-weighted
County type	Weighting	Minus all-county mean (points)	Difference, low	Difference, high	Mean swing minus national swing (points)	Interval low	Interval high	All-county mean, same weighting (points)	Counties	Two-party votes, 2024 (millions)
High-Hispanic	Vote-weighted	−4.2	−6.2	−3.0	−2.9	−4.8	−1.2	+1.3	314	36.6
High-Hispanic	County-equal	−3.3	−4.7	−2.4	−0.6	−2.0	+0.4	+2.7	314	36.6
High-Black	Vote-weighted	+0.9	−0.8	+2.1	+2.2	+0.6	+3.4	+1.3	379	17.5
High-Black	County-equal	−1.8	−2.9	−1.0	+0.9	−0.2	+1.7	+2.7	379	17.5
Affluent suburban	Vote-weighted	+0.8	−0.1	+1.7	+2.1	+0.7	+3.7	+1.3	152	36.2
Affluent suburban	County-equal	+0.3	−0.7	+1.3	+3.0	+1.9	+4.3	+2.7	152	36.2
All other	Vote-weighted	+1.9	+0.8	+2.7	+3.2	+2.6	+3.8	+1.3	2,257	59.7
All other	County-equal	+0.7	+0.4	+1.1	+3.4	+3.0	+3.9	+2.7	2,257	59.7

Bars to the left of zero moved further toward the Republicans than the average county did, and a whisker that crosses zero is a difference whose interval includes zero. Each bar is a group's mean swing minus the national swing, minus the mean over all counties in the same weighting, which is +1.3 vote-weighted and +2.7 county-equal. Whiskers are 95% intervals from resampling states.Note: Intervals resample whole states; see Methods. The groups are rules applied in the order given in the text, and each county is in one group; the thresholds are not fitted to the swing.Note: Counties are assigned by their 2020 characteristics. Jackson County, Missouri and the eight Connecticut counties are left out. The data table also gives each group's mean minus the national swing, with its interval, and the all-county mean.Source: County presidential returns, 2000-2024 (Algara and Amlani; MIT Election Data and Science Lab), full counts of votes, and county demographics (decennial census and American Community Survey estimates interpolated to election years); weighted by two-party votes in the later election or county-equal, as stated

The 314 high-Hispanic counties averaged 4.2 (95% CI 3.0 to 6.2) points further toward the Republicans than the all-county mean vote-weighted, and 3.3 (95% CI 2.4 to 4.7) points further county-equal. The 379 high-Black counties differ from the all-county mean by +0.9 (95% CI −0.8 to 2.1) vote-weighted and −1.8 (95% CI −2.9 to −1.0) county-equal, so their position depends on the weighting. The 152 affluent suburban counties differ by +0.8 (95% CI −0.1 to 1.7) and +0.3 (95% CI −0.7 to 1.3), and both intervals include zero. The county-equal difference for the high-Black counties has the same sign as the county-equal Black-share correlation, −0.24 (95% CI −0.36 to −0.14), and the vote-weighted Black-share correlation is −0.07 (95% CI −0.25 to 0.09).

The high-Hispanic difference stays below the all-county mean when the cut is 15%, at −3.1 (95% CI −4.2 to −2.1) vote-weighted and −2.1 (95% CI −2.9 to −1.3) county-equal. The high-Black difference keeps its signs under each of 6 other cuts from 15% to 50%, from +0.8 to +1.9 vote-weighted and from −2.0 to −1.4 county-equal. The affluent suburban difference is at or above zero vote-weighted under 6 other sets of college, income and rural cuts, from 0.0 to +1.0, and changes sign county-equal, from −0.8 to +0.6.

Kings County, New York, has a Black share of 28.0% and is in the high-Black group, where it has the most votes. Its swing was −5.2 points relative to the nation, so it moved further toward the Republicans than the nation did and lowers the group's vote-weighted mean. Without it the group's mean swing minus the national swing rises from +2.2 to +2.5. Wayne County, Michigan (−2.5) and Essex County, New Jersey (−4.2) are also below the national swing, and without all three the mean is +2.9. Fulton, Cobb, DeKalb, Gwinnett and Henry counties in the Atlanta area are all above the national swing, the smallest at +3.4 points, and without them the mean is +1.8. Of the ten high-Black counties with the most votes, 2 are below the national swing, Kings and Wayne. The group's vote-weighted mean combines large counties on both sides of the national swing.

The Democratic margin fell among Hispanic and Black respondents

The county correlations do not say which voters moved. The Cooperative Election Study (CES) asks respondents how they voted and how they describe their race, and Figure 8 reports the change in the Democratic margin among respondents who reported a major-party vote, using survey weights. Each wave is a separate sample, so the figure compares two groups of respondents and does not follow the same people.The intervals use the Kish effective sample size [4], treat the weighted sample as a simple random sample, and combine the standard errors of the two waves because they are separate samples of respondents. The CES is an opt-in panel and not a probability sample, so the intervals describe sampling variation among its respondents and leave out how those respondents came to join it. The default weight is the year-specific weight, the only one present in all four waves. It is the weight of the pre-election wave, applied here to a post-election vote question. Among Hispanic respondents the margin went from +32.5 in 2020 to +9.5 in 2024, a change of −23.0 (95% CI −30.5 to −15.5) points. Among Black respondents it went from +82.9 to +63.8, a change of −19.0 (95% CI −24.1 to −14.0). Among white respondents the change was −7.0 (95% CI −9.1 to −4.8) in 2012 to 2016, +2.2 (95% CI 0.3 to 4.2) in 2016 to 2020 and −1.9 (95% CI −4.0 to 0.2) in 2020 to 2024, and among Black respondents it was −6.5 (95% CI −9.9 to −3.1) in 2012 to 2016.

Figure 8

The Democratic margin fell among Hispanic and Black respondents in 2020 to 2024 and among white and Black respondents in 2012 to 2016

Change in the Democratic minus Republican margin among White, Black and Hispanic respondents who reported a major-party presidential vote, in percentage points, with one panel for each pair of elections

2012–16

2016–20

2020–24

Change in Democratic minus Republican margin (points)

Show data

Data for The Democratic margin fell among Hispanic and Black respondents in 2020 to 2024 and among white and Black respondents in 2012 to 2016
Race	Pair of elections	Weight	Change in margin (points)	Interval low	Interval high	Respondents, both waves combined
White	2012–16	Year-specific weight	−7.0	−9.1	−4.8	61,083
Black	2012–16	Year-specific weight	−6.5	−9.9	−3.1	6,938
Hispanic	2012–16	Year-specific weight	+4.3	−3.6	+12.3	6,340
Asian	2012–16	Year-specific weight	+11.0	−3.0	+25.0	1,502
Other	2012–16	Year-specific weight	−7.1	−17.9	+3.7	2,817
White	2016–20	Year-specific weight	+2.2	+0.3	+4.2	63,584
Black	2016–20	Year-specific weight	−2.4	−6.1	+1.4	7,967
Hispanic	2016–20	Year-specific weight	−5.7	−12.9	+1.5	7,866
Asian	2016–20	Year-specific weight	−12.5	−22.0	−3.1	2,208
Other	2016–20	Year-specific weight	−2.5	−12.3	+7.3	3,138
White	2020–24	Year-specific weight	−1.9	−4.0	+0.2	64,714
Black	2020–24	Year-specific weight	−19.0	−24.1	−14.0	7,822
Hispanic	2020–24	Year-specific weight	−23.0	−30.5	−15.5	7,642
Asian	2020–24	Year-specific weight	−9.8	−21.1	+1.4	2,210
Other	2020–24	Year-specific weight	−6.7	−17.2	+3.9	3,492
White	2012–16	Post-election weight, validated-voter weight for 2020 to 2024	−4.4	−6.7	−2.2	61,083
Black	2012–16	Post-election weight, validated-voter weight for 2020 to 2024	−6.8	−10.5	−3.0	6,938
Hispanic	2012–16	Post-election weight, validated-voter weight for 2020 to 2024	+10.1	+1.9	+18.3	6,340
Asian	2012–16	Post-election weight, validated-voter weight for 2020 to 2024	+11.0	−3.3	+25.4	1,502
Other	2012–16	Post-election weight, validated-voter weight for 2020 to 2024	−2.6	−13.7	+8.4	2,817
White	2016–20	Post-election weight, validated-voter weight for 2020 to 2024	+3.6	+1.5	+5.7	63,583
Black	2016–20	Post-election weight, validated-voter weight for 2020 to 2024	−4.3	−8.6	+0.1	7,967
Hispanic	2016–20	Post-election weight, validated-voter weight for 2020 to 2024	−4.7	−12.3	+3.0	7,866
Asian	2016–20	Post-election weight, validated-voter weight for 2020 to 2024	−11.5	−22.0	−1.0	2,208
Other	2016–20	Post-election weight, validated-voter weight for 2020 to 2024	+4.7	−6.5	+15.8	3,138
White	2020–24	Post-election weight, validated-voter weight for 2020 to 2024	−4.1	−6.2	−1.9	53,387
Black	2020–24	Post-election weight, validated-voter weight for 2020 to 2024	−13.9	−19.5	−8.3	5,828
Hispanic	2020–24	Post-election weight, validated-voter weight for 2020 to 2024	−14.9	−24.1	−5.6	5,294
Asian	2020–24	Post-election weight, validated-voter weight for 2020 to 2024	−3.1	−15.6	+9.4	1,408
Other	2020–24	Post-election weight, validated-voter weight for 2020 to 2024	+8.1	−4.0	+20.2	2,777

Dots to the left of zero are a fall in the Democratic margin. Among Hispanic respondents the margin changed by −23.0 points in 2020 to 2024, among Black respondents by −19.0 and among white respondents by −1.9; in 2012 to 2016 it changed by −7.0 among white respondents and −6.5 among Black respondents. Each dot is the difference between two waves of the same survey, and the respondents in the two waves are different people. The default view uses the year-specific weight and the second view shows how much the estimates depend on the weight. Asian and other respondents are in the data table and not plotted, because their intervals are wide.Note: The Cooperative Election Study is an opt-in online panel with post-stratification weights, so the intervals use the Kish effective sample size and assume a weighted simple random sample. They describe sampling variation among its respondents and leave out how those respondents came to join the panel. The year-specific weight is the pre-election weight, applied here to a post-election vote question, and it is the only weight present in all four waves.Note: Respondents are those who reported voting for the Democratic or Republican presidential candidate, so people who did not vote and people who voted for others are left out. Race is self-reported. A dot compares two separate samples of respondents, so it does not follow the same people over time.Source: Cooperative Election Study, 2012, 2016, 2020 and 2024 waves; weighted by the year-specific weight, and in the second view by the post-election weight (2012 to 2016 and 2016 to 2020) or the validated-voter weight (2020 to 2024)

The size of these changes depends on the weight. With the validated-voter weight, which covers respondents matched to a voter file, the 2020 to 2024 changes are −14.9 (95% CI −24.1 to −5.6) among Hispanic respondents, −13.9 (95% CI −19.5 to −8.3) among Black respondents and −4.1 (95% CI −6.2 to −1.9) among white respondents. With the post-election weight, the 2012 to 2016 changes are −4.4 (95% CI −6.7 to −2.2) among white respondents and −6.8 (95% CI −10.5 to −3.0) among Black respondents. The Hispanic and Black margins fall in 2020 to 2024 under both weights, and the white change in 2020 to 2024 has an interval that includes zero under the year-specific weight and excludes it under the validated-voter weight.

The Hispanic result agrees in sign with the county correlation for 2020 to 2024. The Black margin fell by −19.0 points among respondents while county Black share has a vote-weighted correlation of −0.07 with the swing, so a county correlation does not measure how a group voted. The Chapter 3.4, Hispanic, Black and Asian voters (planned) chapter takes up the votes of each group.

What these data cannot show

Every correlation above is a property of counties, and a pattern across places can differ from the pattern among the people who live in them, a limit known as ecological inference. Robinson [5] showed that a correlation across places can have the opposite sign from the correlation across individuals. The foreign-born population includes residents who are not citizens and cannot vote, so the foreign-born correlation says where the swing was large and not how immigrants voted. The characteristics describe residents, not voters, and a swing combines people who changed their vote with people who began or stopped voting or moved in or out. Chapter 15.5, Composition and behavior in county change (planned) separates composition from behavior with other data.

The dates are the dates of pairs. A correlation for 2012 to 2016 and one for 2016 to 2020 cannot say when between the two elections a county moved, and a change of sign between two pairs says that the ordering of counties changed, not why. The four characteristics are correlated with each other and with others that are not in the regression, among them income, age and rurality, which Chapter 12.3, Income and the county vote (planned) and Chapter 12.5, Rural counties (planned) take up, so a coefficient describes one regression and not a separate effect. The characteristics are census and American Community Survey estimates interpolated between census and survey years, so they carry sampling error, most in small counties, that no interval here includes. The county set is not harmonized across boundary changes, and the county groups are threshold rules whose group means move with the cut. A swing is not adjusted for the county's earlier margin, and a county that began with an almost one-sided margin had less room to move in one direction than a competitive county did.

The intervals describe dependence on which states are in the set, and the returns themselves are counts. The intervals on group means are of the same kind. The ZIP code check covers only 2016 to 2020, and its votes are allocated from precincts. The CES is an opt-in online panel with self-reported vote, so its estimates depend on how well the weights correct for who joins and on which weight is used. The data used here hold no measure of turnout by group, of attitudes or of campaign activity, so they cannot say why the swing lined up with Hispanic and foreign-born share in 2020 to 2024 and with other characteristics in 2012 to 2016.

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); county swings weighted by two-party votes in the later 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, Hispanic, foreign-born and Black shares, income and rural share at the earlier election of each pairYears: 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.
Precinct votes on census geographiesPrecinct votes allocated to ZIP code areas, 2016 and 2020, with the census atlas measuresYears: 2016 and 2020 (contiguous United States only) · Weights: 2020 presidential votes; areas with 100 votes or fewer dropped · License: CC0 (data doi:10.7910/DVN/Z8TSH3), as stated in the project README.
Cooperative Election StudyCooperative Election Study cumulative file, 2012, 2016, 2020 and 2024 wavesYears: 2006-2024 (every year; self-reported turnout is available in the 10 even years) · Weights: weight (year-specific); sensitivity: weight_post and vvweight · License: See the provider's terms of use.

All data sources

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. A pair of consecutive elections uses the counties present in both, matched on FIPS code, which gives 3,111 to 3,114 counties; county boundaries are not harmonized. County characteristics come from the county demographics panel (decennial census and American Community Survey estimates, interpolated to election years, so they carry sampling error that no interval here includes) at the earlier election of the pair: the share of adults 25 and older with a bachelor's degree or more, the Hispanic share of residents (any race), the foreign-born share of residents and the non-Hispanic Black share of residents; the county groups also use median household income in 2024 dollars and the rural share of residents. Eight Connecticut counties have no characteristics for 2020, and Jackson County, Missouri loses its Kansas City votes to a separate unit in 2024; both are left out of the 2020 to 2024 correlations, regressions, groups and scatter and stay in the national swing, so that the swing matches chapter 10.1; Alaska is not in any county statistic. Estimand: the margin is 100 times the Democratic minus the Republican votes over their sum; a county's swing is its margin in the later election minus its margin in the earlier one; the national swing is the same change in the summed margin of the matched counties; the relative swing is the county swing minus the national swing. A correlation with the swing equals the correlation with the relative swing. Weights: two-party votes in the later election (vote-weighted) or one per county (county-equal). Correlations are Pearson correlations across counties. Regressions are weighted least squares of the swing on the four characteristics, each variable standardized with the weighted mean and standard deviation of the weighting used; non-Hispanic white share is not a regressor. A second regression uses Hispanic and foreign-born share only. R-squared is the share of the variance of the standardized swing that the fitted values reproduce, from the weighted fit. Intervals: 95% percentile intervals from a bootstrap over whole states for correlations, for the mean swing of county groups and for the difference between a group mean and the all-county mean (2,000 replicates, seed 12345, 48 states and the District of Columbia in the 2020 to 2024 pair; for a group mean the states are resampled from all states in the analysis, so a group can gain or lose states), and normal intervals with errors clustered by state for regression coefficients. The returns are full counts, so these intervals show how much a number depends on which states are in the set; they are not sampling error of the vote and they do not include the sampling error of the county characteristics. County groups: high-Hispanic (25% or more Hispanic in 2020), high-Black (25% or more non-Hispanic Black and not high-Hispanic), affluent suburban (35% or more of adults with a bachelor's degree, median household income of $75,000 or more, rural share below 25%, and neither of the first two) and all other counties. The county-type figure plots each group's mean relative swing minus the mean over all counties in the analysis, in the same weighting, because the all-county mean is not zero (the national swing is the change in the summed margin, which large counties dominate). Threshold check: the difference from the all-county mean is recomputed for the high-Hispanic group with a 15% cut, for the high-Black group with six other cuts (15, 20, 30, 35, 40 and 50%), and for the affluent suburban group with six other sets of cuts that move the college cut (to 30 or 40%), the income cut (to $60,000 or $90,000) or the rural cut (to 10 or 50%) one at a time from the baseline. Group means of the swing by Hispanic share (under 10, 10 to under 25, 25 to under 50 and 50% or more) and by college share (under 20, 20 to under 40 and 40% or more) are drawn at the group's median share, the same for both weightings. Large counties: 250,000 or more two-party votes in the later election; correlations inside a size class count each county once, and the table of size classes also gives the mean college share and mean swing of each class, and for all counties in each weighting. ZIP code areas: the 2016 to 2020 swing surface from precinct votes allocated to ZIP code areas (Fekrazad 2025; the swing is the 2020 margin minus the 2016 margin, in points), areas with more than 100 presidential votes in 2020 (29,323 areas), weighted by 2020 presidential votes; its college measure is the atlas's college-education share, which the atlas file does not define and is not the county bachelor's degree share; areas with a negative sentinel in any of the four measures are dropped. Survey check: Cooperative Election Study respondents who reported a Democratic or Republican presidential vote in 2012, 2016, 2020 and 2024, the year-specific weight (the pre-election weight, the only weight present in all four waves, applied to a post-election vote item), the change in the weighted margin between two waves by self-reported race (the waves are separate samples of respondents; the chart shows White, Black and Hispanic respondents and the data table also has Asian and other respondents), with intervals from the Kish effective sample size (two independent waves, standard errors combined in quadrature). Sensitivity to the weight: the post-election weight for 2012 to 2016 and 2016 to 2020 (not in the file for 2024) and the validated-voter weight for 2020 to 2024 (respondents matched to a voter file). Suppression: none for county or ZIP code area figures; a survey cell needs at least 100 respondents and an effective sample of at least 100, and every race-by-wave cell meets both.

References

Sides, John; Tesler, Michael; Vavreck, Lynn (2018). Identity Crisis: The 2016 Presidential Campaign and the Battle for the Meaning of America. Princeton University Press. press.princeton.edu/books/hardcover/9780691174198/identity-crisis
Zingher, Joshua N. (2022). Diploma Divide: Educational Attainment and the Realignment of the American Electorate. Political Research Quarterly 75(2): 263-277. doi:10.1177/10659129221079862
Fekrazad, Amir (2025). A dataset of US precinct votes allocated to Census geographies with precision. Scientific Data 12: 794. doi:10.1038/s41597-025-05140-3
Kish, Leslie (1965). Survey Sampling. John Wiley and Sons.
Robinson, W. S. (1950). Ecological Correlations and the Behavior of Individuals. American Sociological Review 15(3): 351-357.

How to cite

@misc{singh2026educationethnicityandthecountyshift,
  author = {Singh, Vishal},
  title = {Education, ethnicity and the county shift, 2016 to 2024},
  year = {2026},
  url = {https://vishalsingh.org/politics/education-ethnicity-and-the-county-shift},
  note = {In The Public Mind. Published 30 September 2026.}
}

Same question at another grain

These chapters take up the same axis with a different unit of analysis: individuals, places or countries.

Race and ethnicity
- Chapter 3.4, Hispanic, Black and Asian voters (planned) Individuals
- Chapter 12.5, Rural counties (planned) Places

Vote-weighted Hispanic and foreign-born correlations turned negative after 2016

Counties furthest below the national swing had high Hispanic shares

The college fall depends on the weighting

The Hispanic gradient steepens at high shares

Holding other shares fixed, college share re-emerges

ZIP code areas reproduce the Hispanic and foreign-born correlations

High-Hispanic counties swung furthest toward the Republicans

The Democratic margin fell among Hispanic and Black respondents

What these data cannot show

Data sources

Methods

References

How to cite

Same question at another grain

Reading paths