Chapter 12.4

Wages, capital and the ZIP code vote

ZIP codes with more capital-gains returns swung more toward the Democrats between 2016 and 2020 and ZIP codes with more EITC returns swung less, but the capital-gains sign depends on the controls and the EITC link is near zero with poverty, college and race held fixed, and the definition of the ZIP-code college variable is unresolved.

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

DraftFigures and numbers may change as the analysis is rebuilt.what is still being checked

The ACS release behind the ZIP-code controls (poverty, college, Black and Hispanic shares) is not recorded in the demographic source file. The ZIP-code controls are to be rebuilt from a dated 5-year ACS release; the county results use dated controls and give the same conclusions for the EITC share. The ZIP-code college variable has no resolved definition (some college or more, or degrees), so every ZIP-code result that holds it fixed is conditional on that coding.

ZIP codes in the highest decile of the share of returns that claim the earned income tax credit (EITC) swung 5.2 points toward the Republicans between 2016 and 2020, and ZIP codes in the highest decile of the share of returns that report a capital gain swung 5.4 points toward the Democrats. The coefficient of the swing on the capital-gains share, in points of margin per standard deviation of the share, is +2.4 (95% CI 2.0 to 2.9) with no controls and −1.0 (95% CI −1.6 to −0.4) with the college share held fixed. To hold the college share fixed is to compare ZIP codes with similar college shares.A coefficient here is the difference in swing, in points of margin, between ZIP codes one standard deviation apart on the measure. One standard deviation is 11.5 points of share for capital gains and 8.3 for the EITC. The EITC coefficient goes from −3.5 (95% CI −4.0 to −3.0) with no controls to +0.3 (95% CI −0.8 to 1.4) with poverty, college, Black and Hispanic shares held fixed, a set that leaves out the average tax rate. The capital-gains share correlates +0.81 (95% CI 0.79 to 0.82) with the college share, so these data cannot separate the two.

The unit is the ZIP code tabulation area, the census approximation of a ZIP code, and not the county. The sample is 27,144 ZIP codes in 3,088 counties, and the county analysis uses 3,080 counties; Figure 8 in the last section lists the exclusions. The EITC share stands in for low wages, as the next section explains. Counties enter in two ways: as the clusters that set the width of the ZIP-code intervals, and as a second grain in which ZIP-code shares are averaged to counties and compared with certified county returns. Unless the text says otherwise, ZIP-code statistics weight each ZIP code by its 2020 votes and county statistics weight each county by its 2020 two-party votes.

What the shares and the swing measure

The earned income tax credit is a refundable federal credit for households with low earnings from work, so the share of a ZIP code's returns that claim it is higher where more filers have low earnings; it also depends on children in the household and on take-up. A capital gain is the profit from selling an asset such as stock or a home, and returns also report gains that funds distribute, so the share of returns that report one is higher where more households hold assets that produce gains. Both shares come from IRS Statistics of Income data for tax year 2022, and both describe filers: households that do not file a return are absent. The vote-weighted mean is 14.2% of returns for the EITC and 20.1% for capital gains, and across ZIP codes the two shares correlate −0.78 (95% CI −0.79 to −0.77) (−0.68 with each ZIP code counted equally), so a ZIP code with many EITC claims has few capital-gains returns and the reverse.The IRS shares are for tax year 2022, which follows both elections they are set against. The ZIP-code demographic file that carries the poverty, college and race controls does not record which American Community Survey release it uses, so those controls are undated. The county controls come from a panel with documented anchor years. The last section returns to the timing.

Votes for 2016 and 2020 are precinct returns that Fekrazad allocated to ZIP code tabulation areas in proportion to household population [1], from the Voting and Election Science Team's 2016 precinct returns [2] and the 2020 returns of the MIT Election Data and Science Lab [3]. The outcome is the swing: the 2020 two-party margin minus the 2016 margin, Democratic minus Republican, in points, so a positive swing is toward the Democrats. Across the ZIP codes in this chapter the vote-weighted mean of the swing is 3.0 points toward the Democrats.The mean weights both years' margins of a ZIP code by its 2020 votes. The aggregate two-party margin of the same ZIP codes weights each year by that year's votes and moved 2.3 points, from +2.3 to +4.6.

Swing rose with capital gains and fell with EITC

Figure 1 ranks the ZIP code tabulation areas by each share and splits them into ten equal groups, or deciles. The mean swing is 7.2 (95% CI 5.6 to 8.8) points toward the Republicans in the lowest capital-gains decile and 5.4 (95% CI 4.4 to 6.4) points toward the Democrats in the highest, and it is 6.3 (95% CI 5.5 to 7.2) points toward the Democrats in the lowest EITC decile and 5.2 (95% CI 3.5 to 6.9) points toward the Republicans in the highest. Most deciles of both shares swung toward the Democrats, as the mean of all ZIP codes did, so the ordering is one of how far the deciles moved. Intervals use clustered standard errors at the county, because ZIP codes in one county share local conditions and are not independent draws.

Figure 1

The swing rose with the capital-gains share and fell with the EITC share

Vote-weighted mean change in the two-party margin, 2016 to 2020, by decile of ZIP codes ranked by share of returns; 95% intervals clustered on county; positive is toward the Democrats

Change in Democratic minus Republican margin (points)

Show data

Data for The swing rose with the capital-gains share and fell with the EITC share
Share ranked	Decile	Lowest share in decile (%)	Highest share in decile (%)	Mean swing (points)	Lower bound, county clusters	Upper bound, county clusters	Lower bound, state clusters	Upper bound, state clusters	ZIP codes
Capital-gains share	1	0.0	5.5	−7.2	−8.8	−5.6	−9.9	−4.5	2,715
Capital-gains share	2	5.5	8.9	−1.9	−3.5	−0.4	−4.2	+0.3	2,714
Capital-gains share	3	8.9	11.3	+0.9	−0.1	+1.8	−0.8	+2.6	2,714
Capital-gains share	4	11.3	13.5	+1.9	+1.2	+2.7	+0.7	+3.1	2,715
Capital-gains share	5	13.5	15.7	+3.2	+2.5	+4.0	+1.8	+4.7	2,714
Capital-gains share	6	15.7	18.1	+3.9	+3.1	+4.6	+2.4	+5.3	2,714
Capital-gains share	7	18.1	20.9	+4.5	+3.9	+5.2	+3.4	+5.6	2,715
Capital-gains share	8	20.9	24.7	+5.4	+4.8	+6.0	+4.3	+6.5	2,714
Capital-gains share	9	24.7	31.0	+6.1	+5.4	+6.8	+4.6	+7.6	2,714
Capital-gains share	10	31.0	77.1	+5.4	+4.4	+6.4	+3.4	+7.4	2,715
EITC share	1	0.0	5.1	+6.3	+5.5	+7.2	+4.7	+8.0	2,715
EITC share	2	5.1	7.7	+6.2	+5.5	+6.8	+4.6	+7.8	2,714
EITC share	3	7.7	9.7	+5.7	+5.0	+6.4	+4.2	+7.2	2,714
EITC share	4	9.7	11.5	+4.9	+4.2	+5.5	+3.6	+6.1	2,715
EITC share	5	11.5	13.4	+4.3	+3.6	+4.9	+3.0	+5.5	2,714
EITC share	6	13.4	15.4	+3.3	+2.5	+4.1	+1.7	+4.8	2,714
EITC share	7	15.4	18.0	+2.0	+0.7	+3.2	−0.2	+4.2	2,715
EITC share	8	18.0	21.3	+0.8	−0.4	+2.0	−1.2	+2.8	2,714
EITC share	9	21.3	26.3	−1.5	−3.2	+0.1	−4.0	+1.0	2,714
EITC share	10	26.3	72.7	−5.2	−6.9	−3.5	−8.1	−2.3	2,715

The two lines cross between the fifth and sixth deciles. Capital-gains ZIP codes went from a Republican swing at the bottom (7.2 points) to a Democratic one at the top (5.4); EITC ZIP codes did the opposite (6.3 Democratic to 5.2 Republican). Most deciles of both shares swung toward the Democrats, as the mean of all ZIP codes did (3.0 points).Note: ZIP code tabulation areas are ranked by the share of their returns that report a net capital gain or claim the earned income tax credit and split into ten groups of equal size. Each point is the mean swing weighted by 2020 votes, and the band is a 95% interval clustered on county. The data table also gives intervals clustered on state, which are wider; the intervals of the four end deciles still exclude zero. ZIP codes with 100 or fewer votes in 2020 are excluded.Source: Precinct votes allocated to ZIP code tabulation areas (Fekrazad 2025), 2016 and 2020; IRS Statistics of Income ZIP-code data, tax year 2022; ACS 5-year estimates; weighted by 2020 votes

Figure 2 plots the 2020 margin for the same deciles. It bends into a U for both shares: the margin was most Democratic in the lowest and highest deciles and lowest in the middle of the range.

Figure 2

The 2020 margin was most Democratic at both ends of each share and lowest in the middle of the range

Vote-weighted mean 2020 two-party margin, Democratic minus Republican, by decile of ZIP codes ranked by share of returns; 95% intervals clustered on county

Democratic minus Republican margin, 2020 (points)

Show data

Data for The 2020 margin was most Democratic at both ends of each share and lowest in the middle of the range
Share ranked	Decile	Mean margin, 2020 (points)	Lower bound	Upper bound	ZIP codes
Capital-gains share	1	+48.4	+42.5	+54.3	2,715
Capital-gains share	2	+14.0	+8.8	+19.2	2,714
Capital-gains share	3	−4.8	−9.6	0.0	2,714
Capital-gains share	4	−11.6	−15.7	−7.6	2,715
Capital-gains share	5	−12.0	−15.2	−8.9	2,714
Capital-gains share	6	−6.9	−10.3	−3.4	2,714
Capital-gains share	7	−6.9	−10.2	−3.6	2,715
Capital-gains share	8	+0.8	−2.7	+4.3	2,714
Capital-gains share	9	+5.6	+2.1	+9.0	2,714
Capital-gains share	10	+22.6	+17.3	+27.9	2,715
EITC share	1	+19.9	+13.5	+26.4	2,715
EITC share	2	+9.5	+5.7	+13.4	2,714
EITC share	3	+2.5	−1.2	+6.2	2,714
EITC share	4	−0.8	−4.8	+3.1	2,715
EITC share	5	−6.0	−9.6	−2.5	2,714
EITC share	6	−6.3	−10.5	−2.0	2,714
EITC share	7	−6.1	−11.9	−0.2	2,715
EITC share	8	−3.7	−9.1	+1.8	2,714
EITC share	9	+3.4	−3.0	+9.9	2,714
EITC share	10	+33.3	+27.3	+39.2	2,715

Both lines bend into a U, with the lowest margins in the middle deciles. The ends of the capital-gains line are +48.4 and +22.6 points and the ends of the EITC line +19.9 and +33.3, against −12.0 and −6.3 at the low points. Several middle intervals include zero, so the figure shows a dip and not a Republican margin in every middle decile.Note: Deciles, weights and intervals are as in the swing figure; the vertical axis is the 2020 margin and not the change.Source: Precinct votes allocated to ZIP code tabulation areas (Fekrazad 2025), 2016 and 2020; IRS Statistics of Income ZIP-code data, tax year 2022; ACS 5-year estimates; weighted by 2020 votes

The two ends of each U are different kinds of places (Figure 3). The lowest capital-gains decile and the highest EITC decile are poor ZIP codes with large Black and Hispanic populations, and they are the two end deciles that swung toward the Republicans. The other two ends, the highest capital-gains decile and the lowest EITC decile, are college-heavy. A ZIP code can be solidly Democratic in level and still have moved toward the Republicans, so the swing orders the deciles from one end of a share to the other while the level is high at both ends.

Figure 3

The two end deciles that swung toward the Republicans are the poorest of the four

Vote-weighted mean swing, 2020 margin and demographic profile of the lowest and highest deciles of each share; positive swing and margin are toward the Democrats

ZIP codes in the	Swing, 2016 to 2020 (points)	Margin, 2020 (points)	Black and Hispanic share (%)	Poverty rate (%)	ZIP-code college variable (%)
Lowest capital-gains decile	−7.2	+48.4	76.5	23.4	40.6
Highest capital-gains decile	+5.4	+22.6	12.8	3.8	82.1
Lowest EITC decile	+6.3	+19.9	8.5	2.7	82.9
Highest EITC decile	−5.2	+33.3	66.8	22.9	43.9

Show data

Data for The two end deciles that swung toward the Republicans are the poorest of the four
ZIP codes in the	Swing (points)	Lower bound	Upper bound	Margin, 2020 (points)	Lower bound	Upper bound	Black and Hispanic share (%)	Lower bound	Upper bound	Poverty rate (%)	Lower bound	Upper bound	ZIP-code college variable (%)	Lower bound	Upper bound	ZIP codes
Lowest capital-gains decile	−7.2	−8.8	−5.6	+48.4	+42.5	+54.3	76.5	72.8	80.2	23.4	22.4	24.4	40.6	38.9	42.3	2,715
Highest capital-gains decile	+5.4	+4.4	+6.4	+22.6	+17.3	+27.9	12.8	11.7	13.8	3.8	3.6	4.0	82.1	81.3	82.8	2,715
Lowest EITC decile	+6.3	+5.5	+7.2	+19.9	+13.5	+26.4	8.5	7.9	9.1	2.7	2.6	2.8	82.9	81.8	83.9	2,715
Highest EITC decile	−5.2	−6.9	−3.5	+33.3	+27.3	+39.2	66.8	63.2	70.5	22.9	22.2	23.7	43.9	42.9	45.0	2,715

The two ends with a Republican swing have poverty rates of 23.4% and 22.9% and Black and Hispanic shares of 76.5% and 66.8%, against 9.2% and 26.4% across the ZIP codes in the sample. The two ends with a Democratic swing have the largest college shares: 82.1% and 82.9%, against 40.6% in the lowest capital-gains decile.Note: The Black and Hispanic shares are added, and a Black Hispanic resident counts in both, so the sum is an upper bound on the share of residents in either group. Means are weighted by 2020 votes; the data table gives 95% intervals clustered on county.Source: Precinct votes allocated to ZIP code tabulation areas (Fekrazad 2025), 2016 and 2020; IRS Statistics of Income ZIP-code data, tax year 2022; ACS 5-year estimates; weighted by 2020 votes

Holding college fixed reverses the capital-gains sign

In these data a ZIP code's college share is the college variable of the ZIP-code demographic file, and what that variable counts is unresolved.The file describes the variable as the percentage of the population with a college degree or higher and names no table, age group or release. Aggregated to counties it averages 52.7% against 23.6% for the county bachelor's share, with a correlation of +0.86, so its level is far above a bachelor's share. The level and the correlation do not establish whether it counts some college or more or degrees only, so every result below that holds it fixed is conditional on that coding. The county analysis uses the bachelor's share. It correlates +0.81 (95% CI 0.79 to 0.82) with the capital-gains share and +0.47 (95% CI 0.43 to 0.51) with the swing (+0.44 with each ZIP code counted equally). A standard deviation of the college share goes with +3.4 (95% CI 3.1 to 3.8) points of swing, against +2.4 (95% CI 2.0 to 2.9) for a standard deviation of the capital-gains share.

Figure 4 adds controls to each share one set at a time, as separate models on the same 27,144 ZIP codes. With the college share held fixed (third row), the capital-gains coefficient goes from +2.4 (95% CI 2.0 to 2.9) with no controls to −1.0 (95% CI −1.6 to −0.4). Holding the poverty rate fixed instead brings it to +0.3 (95% CI −0.1 to 0.7), so most of the raw capital-gains association is shared with low poverty as well as with college. With intervals clustered on state (49 clusters) instead of county, the capital-gains intervals with the college share held fixed, alone (−1.0 (95% CI −2.1 to 0.0)) or with the average tax rate (+1.4 (95% CI −0.3 to 3.0)), include zero, and every other interval in the ladder keeps its position relative to zero. The same check on the deciles leaves the intervals of the four end deciles away from zero.

Figure 4

Controls take the EITC coefficient to about zero and change the sign of the capital-gains coefficient

Weighted regression coefficient of the 2016 to 2020 swing on the share of returns, one set of controls per row; 95% intervals clustered on county in the first view and on state in the second

EITC share

Capital-gains share

Points of swing per standard deviation of the share (positive is toward the Democrats)

Show data

Data for Controls take the EITC coefficient to about zero and change the sign of the capital-gains coefficient
Share	Controls	Coefficient	Lower bound, county clusters	Upper bound, county clusters	Interval, county clusters	Lower bound, state clusters	Upper bound, state clusters	Interval, state clusters	Partial correlation	Lower bound	Upper bound	ZIP codes	County clusters	State clusters
EITC share	No controls	−3.48	−4.01	−2.96	Below zero	−4.31	−2.65	Below zero	−0.475	−0.520	−0.431	27,144	3,088	49
EITC share	Poverty rate	−1.69	−2.73	−0.65	Below zero	−3.35	−0.03	Below zero	−0.140	−0.221	−0.068	27,144	3,088	49
EITC share	Average tax rate	−4.12	−5.00	−3.23	Below zero	−5.50	−2.74	Below zero	−0.428	−0.488	−0.371	27,144	3,088	49
EITC share	College share	−2.08	−2.86	−1.30	Below zero	−3.38	−0.78	Below zero	−0.222	−0.289	−0.158	27,144	3,088	49
EITC share	Average tax rate and college share	−2.76	−3.58	−1.93	Below zero	−4.10	−1.41	Below zero	−0.290	−0.356	−0.229	27,144	3,088	49
EITC share	Poverty, college, Black and Hispanic shares	+0.32	−0.76	+1.39	Includes zero	−1.24	+1.87	Includes zero	+0.021	−0.051	+0.085	27,144	3,088	49
Capital-gains share	No controls	+2.42	+1.98	+2.86	Above zero	+1.84	+3.00	Above zero	+0.330	+0.288	+0.376	27,144	3,088	49
Capital-gains share	Poverty rate	+0.31	−0.09	+0.71	Includes zero	−0.23	+0.85	Includes zero	+0.038	−0.006	+0.087	27,144	3,088	49
Capital-gains share	Average tax rate	+4.05	+2.59	+5.51	Above zero	+2.58	+5.51	Above zero	+0.249	+0.181	+0.314	27,144	3,088	49
Capital-gains share	College share	−1.02	−1.61	−0.43	Below zero	−2.08	+0.03	Includes zero	−0.093	−0.145	−0.038	27,144	3,088	49
Capital-gains share	Average tax rate and college share	+1.38	+0.13	+2.62	Above zero	−0.26	+3.02	Includes zero	+0.087	+0.017	+0.155	27,144	3,088	49
Capital-gains share	Poverty, college, Black and Hispanic shares	−2.74	−3.22	−2.26	Below zero	−3.49	−2.00	Below zero	−0.258	−0.296	−0.214	27,144	3,088	49

The EITC coefficient runs from −3.5 with no controls to +0.3 with poverty, college and race held fixed, and the capital-gains coefficient changes sign with the choice of controls. A line that crosses the dashed zero rule is an interval that includes zero. Each row is a separate model on the same 27,144 ZIP codes.Note: Each row is a separate regression with the controls named on the row, and each uses the same ZIP codes. Race means the Black share and the Hispanic share of residents. The full set is poverty, college, Black and Hispanic shares and does not include the average tax rate, which is the mean effective federal income tax rate in the ZIP code. The ZIP-code college share is the file's college variable, whose definition is unresolved: the file describes it as the share with a college degree or higher, but its level is far above a bachelor's share. Every row that holds it fixed depends on what it counts. The row that holds only the average tax rate fixed is in the data table and not in the chart.Note: Partial correlations for each row are in the data table.Source: Precinct votes allocated to ZIP code tabulation areas (Fekrazad 2025), 2016 and 2020; IRS Statistics of Income ZIP-code data, tax year 2022; ACS 5-year estimates; weighted by 2020 votes

The negative sign with the college share held fixed does not show a Republican lean among capital-gains filers. The capital-gains share and the college share correlate +0.81, so the regression has little independent variation in the capital-gains share to work with, and the sign depends on which other controls are present: with the average tax rate and the college share held fixed together the coefficient is +1.4 (95% CI 0.1 to 2.6). Work on Western democracies from 1948 to 2020 found that higher-educated voters had moved toward parties of the left while high-income voters continued to vote for the right [4], which is consistent with an education pattern in the ZIP-code swing, although that work describes individual voters and these data describe ZIP codes.

The EITC link is near zero with poverty, college and race fixed

The EITC coefficient in Figure 4 is −3.5 (95% CI −4.0 to −3.0) with no controls. It is −1.7 (95% CI −2.7 to −0.6) with the poverty rate held fixed, −2.1 (95% CI −2.9 to −1.3) with the college share held fixed and −2.8 (95% CI −3.6 to −1.9) with the average tax rate and college share held fixed together. With the poverty rate, college share, Black share and Hispanic share all held fixed, a set that does not include the tax rate, it is +0.3 (95% CI −0.8 to 1.4), an interval that includes zero under both clusterings (+0.3 (95% CI −1.2 to 1.9) with clusters on state). A ZIP code with many EITC claims is, on average, poor, has a lower college share and has more Black and Hispanic residents: the EITC share correlates −0.72 (95% CI −0.74 to −0.71) with the college share (−0.56 with each ZIP code counted equally). The link between the EITC share and the swing is therefore partly a link between those characteristics and the swing, and the data do not show a separate association once they are held fixed. Chapter 13.3, Deprivation and mobility (planned) asks the same question of neighborhood deprivation, with race held fixed.

Starr County, Texas, and Holmes County, Mississippi, illustrate that the EITC share alone does not fix the swing. They have nearly the same share of returns claiming the credit, 45.4% and 45.2%. Between 2016 and 2020 Starr swung 56.3 points toward the Republicans, the largest such swing of any county in the county sample, and Holmes 1.7 points toward the Republicans. The Hispanic share of residents is 98.1% in Starr and the Black share is 82.4% in Holmes, and both shares are in the full set of controls. The pair is picked for the contrast, and the ZIP-code regressions do not test whether these shares account for either swing.

The average tax rate was included as a rough stand-in for income, and it is a poor control. It correlates +0.90 (95% CI 0.89 to 0.91) with the capital-gains share and −0.67 (95% CI −0.69 to −0.65) with the EITC share, so it is highest in the ZIP codes with many capital-gains returns and lowest where EITC claims are common, and holding it fixed leaves 19% of the variance of the capital-gains share. It may itself respond to either share, since the credit reduces the tax a filer owes and gains are taxed under their own rates. The rows that hold it fixed rest on a near-collinear control and are the least reliable in the ladder.

Counties show the same raw signs, with wide intervals

Figure 5 maps both shares and the swing by county. Each county averages the shares of its ZIP codes with weights of 2020 votes and takes the swing from certified county returns. The county correlation of the EITC share with the swing is −0.42 (95% CI −0.58 to −0.23) (−0.44 with each county counted equally) and of the capital-gains share +0.21 (95% CI 0.04 to 0.40) (+0.41), the same signs as in the ZIP codes. This changes the geographic grain and the source of the vote and is not an independent test of the IRS shares, which are the same ZIP-code values averaged. Intervals cluster on state, with 48 clusters (the states and the District of Columbia, without Connecticut), and are wider.

Figure 5

The EITC share is highest in the Mississippi Delta, Alabama's Black Belt and along the Rio Grande

Share of returns claiming the earned income tax credit by county, IRS tax year 2022, averaged from ZIP codes within the county

Loading county values

Share of returns claiming the earned income tax credit (%)

Values outside 0 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.

Counties with a higher EITC share tended to swing less toward the Democrats: the county correlation of the share with the swing is −0.42. High EITC shares include counties in the Mississippi Delta and Alabama's Black Belt, along the Rio Grande in Texas and on reservations in South Dakota. Each county averages the shares of its ZIP codes, weighted by their 2020 votes; the swing comes from certified county returns.Note: Counties with no ZIP-code shares or no county demographics draw as no data, as do Alaska and Hawaii, which have no ZIP-code vote estimates. A county with a share of exactly zero in a view also draws as no data in that view, because a zero cannot be told from a small count rounded down. Values are rounded to one decimal place. A county whose shares come from one small ZIP code can show an extreme value.Source: Certified county returns, 2016 and 2020; IRS Statistics of Income ZIP-code shares, tax year 2022, averaged within counties; county demographics centred on 2020; weighted by 2020 two-party votes

Figure 6 repeats the ZIP-code ladder at county grain with the county bachelor's share as the college control; its second view adds the row that holds only the average tax rate fixed. The capital-gains share correlates +0.83 (95% CI 0.77 to 0.87) with the bachelor's share across counties (+0.68 with each county counted equally). The capital-gains coefficient is +1.2 (95% CI 0.0 to 2.4) with no controls, +0.1 (95% CI −1.7 to 2.0) with the bachelor's share held fixed and −2.9 (95% CI −4.4 to −1.5) with the full set held fixed, and the first two intervals include zero. With the average tax rate and the bachelor's share held fixed together it is +2.7 (95% CI 0.4 to 4.9), an interval above zero. The EITC coefficient is −2.4 (95% CI −3.6 to −1.1) with no controls, −2.5 (95% CI −4.2 to −0.8) with the bachelor's share held fixed and 0.0 (95% CI −2.4 to 2.4) with the poverty rate, bachelor's share, Black share and Hispanic share held fixed, an interval that includes zero. The county intervals are wide. The interval for the capital-gains coefficient with the bachelor's share held fixed includes both zero and the ZIP-code estimate with the college share held fixed (−1.0 (95% CI −2.1 to 0.0) with clusters on state), so the county data cannot say whether the capital-gains coefficient after education is near zero or negative, and they do not show that the ZIP-code and county results differ. Under the full set the coefficient is negative at both grains. Chapter 12.3, Income and the county vote (planned) takes county income directly.

Figure 6

At county grain the intervals are wide and the capital-gains coefficient with college held fixed includes zero

Weighted regression coefficient of the 2016 to 2020 county swing on the county share of returns, one set of controls per row; 95% intervals clustered on state

EITC share

Capital-gains share

Points of swing per standard deviation of the share (positive is toward the Democrats)

Show data

Data for At county grain the intervals are wide and the capital-gains coefficient with college held fixed includes zero
Share	Controls	Coefficient	Lower bound	Upper bound	Interval	Counties	State clusters
EITC share	No controls	−2.39	−3.65	−1.14	Below zero	3,080	48
EITC share	Poverty rate	−0.90	−3.56	+1.76	Includes zero	3,080	48
EITC share	Average tax rate	−3.44	−5.03	−1.84	Below zero	3,080	48
EITC share	College share	−2.53	−4.20	−0.85	Below zero	3,080	48
EITC share	Average tax rate and college share	−3.00	−4.52	−1.49	Below zero	3,080	48
EITC share	Poverty, college, Black and Hispanic shares	+0.01	−2.40	+2.43	Includes zero	3,080	48
Capital-gains share	No controls	+1.18	−0.01	+2.37	Includes zero	3,080	48
Capital-gains share	Poverty rate	−0.82	−2.01	+0.38	Includes zero	3,080	48
Capital-gains share	Average tax rate	+3.67	+1.54	+5.81	Above zero	3,080	48
Capital-gains share	College share	+0.14	−1.72	+1.99	Includes zero	3,080	48
Capital-gains share	Average tax rate and college share	+2.67	+0.40	+4.94	Above zero	3,080	48
Capital-gains share	Poverty, college, Black and Hispanic shares	−2.93	−4.40	−1.46	Below zero	3,080	48

Intervals are clustered on state, so they are wider than at ZIP-code grain. The capital-gains coefficient is +1.2 with no controls and +0.1 with the bachelor's share held fixed, and both intervals include zero; the EITC coefficient is −2.4 with no controls and 0.0 with the full set, an interval that includes zero.Note: Swing is taken from certified county returns and the shares are averaged from ZIP codes within each county. Each row is a separate regression with the controls named on the row. The full set is poverty, bachelor's share, Black and Hispanic shares and does not include the average tax rate. County shares are the means of the ZIP-code shares in the county, weighted by 2020 votes; the college control is the county bachelor's share, which is not the same variable as the ZIP-code college share (definition unresolved). The second view adds the row that holds only the average tax rate fixed.Source: Certified county returns, 2016 and 2020; IRS Statistics of Income ZIP-code shares, tax year 2022, averaged within counties; county demographics centred on 2020; weighted by 2020 two-party votes

Investors and homeowners lean Republican within education groups

The ZIP-code results describe places. The Cooperative Election Study asks respondents whether they have money invested in the stock market and whether they own their home, and it records their vote, so it can compare people with the same education. A college degree here means a four-year degree or more. Figure 7 gives the Democratic minus Republican margin among voters for the two major parties in 2020 and in 2016. The investor comparison rests on 43,785 voters in 2020 and 40,803 in 2016 and the homeowner comparison on 42,142 and 39,458, after setting aside voters without an answer to the item and, for homeowners, voters who report a housing status other than owning or renting.

Among voters without a college degree, investors were 27.1 (95% CI 23.4 to 30.8) points more Republican than non-investors in 2020 and 18.1 (95% CI 14.2 to 21.9) points more Republican in 2016. Among college graduates the gaps were 10.3 (95% CI 6.5 to 14.2) and 5.9 (95% CI 2.2 to 9.6) points. Homeowners were more Republican than renters at both education levels: 46.7 (95% CI 42.5 to 50.9) points among non-graduates and 38.0 (95% CI 34.1 to 41.9) points among graduates in 2020, and 47.4 (95% CI 43.3 to 51.6) and 37.1 (95% CI 33.0 to 41.1) points in 2016.

Figure 7

Investors voted more Republican than non-investors with the same education in 2016 and 2020

Democratic minus Republican margin among voters for the two major parties, by year, education and whether the respondent reports investments; 95% intervals; positive is toward the Democrats

Non-investors
Investors

Show data

Data for Investors voted more Republican than non-investors with the same education in 2016 and 2020
Item	Year and education	Group	Margin (points)	Lower bound	Upper bound	Voters	Comparison group	Margin (points)	Lower bound	Upper bound	Voters	Group minus comparison (points)	Lower bound	Upper bound	Group's change, 2016 to 2020 (points)	Lower bound	Upper bound	Comparison group's change (points)	Lower bound	Upper bound	Group's change minus comparison's (points)	Lower bound	Upper bound
investor	2016, no college degree	Investors	−15.2	−18.2	−12.2	9,191	Non-investors	+2.9	+0.4	+5.3	15,098	−18.1	−21.9	−14.2	n/a	n/a	n/a	n/a	n/a	n/a	n/a	n/a	n/a
investor	2020, no college degree	Investors	−20.7	−23.5	−18.0	9,501	Non-investors	+6.4	+3.8	+8.9	15,452	−27.1	−30.8	−23.4	−5.5	−9.6	−1.5	+3.5	0.0	+7.0	−9.1	−14.4	−3.7
investor	2016, college degree	Investors	+14.9	+12.6	+17.1	10,328	Non-investors	+20.8	+17.8	+23.8	6,186	−5.9	−9.6	−2.2	n/a	n/a	n/a	n/a	n/a	n/a	n/a	n/a	n/a
investor	2020, college degree	Investors	+19.4	+17.3	+21.4	12,826	Non-investors	+29.7	+26.4	+32.9	6,006	−10.3	−14.2	−6.5	+4.5	+1.4	+7.5	+8.9	+4.5	+13.3	−4.4	−9.7	+1.0
homeowner	2016, no college degree	Homeowners	−20.1	−22.3	−17.9	16,702	Renters	+27.3	+23.8	+30.8	6,697	−47.4	−51.6	−43.3	n/a	n/a	n/a	n/a	n/a	n/a	n/a	n/a	n/a
homeowner	2020, no college degree	Homeowners	−20.6	−22.8	−18.4	16,241	Renters	+26.2	+22.6	+29.7	7,650	−46.7	−50.9	−42.5	−0.4	−3.5	+2.7	−1.1	−6.2	+3.9	+0.7	−5.2	+6.6
homeowner	2016, college degree	Homeowners	+7.9	+5.8	+9.9	12,186	Renters	+44.9	+41.5	+48.4	3,873	−37.1	−41.1	−33.0	n/a	n/a	n/a	n/a	n/a	n/a	n/a	n/a	n/a
homeowner	2020, college degree	Homeowners	+12.7	+10.6	+14.7	13,685	Renters	+50.6	+47.3	+54.0	4,566	−38.0	−41.9	−34.1	+4.8	+1.9	+7.7	+5.7	+0.9	+10.5	−0.9	−6.5	+4.7

Within each education group the filled dot for investors lies to the left of the open dot for non-investors, so investors were more Republican in both elections. The number at the right of each row is the investor margin minus the non-investor margin, in points; in 2020 the gap was 27.1 points among voters without a college degree and 10.3 points among graduates.Note: A college degree means a four-year degree or more; some college and two-year degrees count as no college degree. Intervals are model based: they use the effective sample size implied by the weights and treat the weighted sample as a simple random sample. Vote choice is self-reported in the post-election wave; the investor and homeownership items come from the respondent's profile. The 2016 and 2020 estimates come from separate samples of voters, and voters who report a housing status other than owning or renting are left out of the homeowner view. The data table gives each group's change in margin from 2016 to 2020 and the difference between the two changes on the 2020 rows.Source: Cooperative Election Study, 2016 and 2020; respondents who voted for one of the two major parties; weighted (post-election weight)

The survey also allows a comparison with the swing, with a caution: the 2016 and 2020 samples are separate cross-sections, and the groups change. The investor share of graduate voters is 63.7% (95% CI 62.8% to 64.6%) in 2016 and 67.9% (95% CI 67.0% to 68.7%) in 2020, and of other voters 33.8% and 35.1%. Between 2016 and 2020 the margin of investors without a degree moved 5.5 (95% CI 1.5 to 9.6) points toward the Republicans, and the margin of non-investors without a degree moved 3.5 (95% CI 0.0 to 7.0) points toward the Democrats (the interval starts at zero), a difference of 9.1 (95% CI 3.7 to 14.4) points. Among graduates, investors moved 4.5 (95% CI 1.4 to 7.5) points toward the Democrats and non-investors 8.9 (95% CI 4.5 to 13.3), a difference of 4.4 (95% CI −1.0 to 9.7) points, an interval that includes zero. The differences are computed from unrounded moves. Investors without a degree therefore moved toward the Republicans while non-investors without a degree moved toward the Democrats, and the data do not show a difference among graduates. Homeowners without a degree moved 0.7 (95% CI −5.2 to 6.6) points more toward the Democrats than renters without a degree, and homeowners with a degree moved 0.9 (95% CI −4.7 to 6.5) points less than renters with a degree. Both homeowner intervals include zero and are wide, so the data do not show whether homeowners and renters moved alike.

The individual pattern is consistent with the same composition as the ZIP-code pattern. College graduates are more likely to invest, 67.9% (95% CI 67.0% to 68.7%) of graduate voters in 2020 against 35.1% (95% CI 34.2% to 36.0%) of others, and graduates vote more Democratic than non-graduates within every investor and housing group in the figure. A ZIP code with many college-educated adults has many capital-gains returns and moved toward the Democrats, while an investor with a given degree leans Republican relative to a non-investor with the same degree. The ZIP-code college variable has an unresolved definition and the survey measure is a four-year degree, so the two may not be the same split. Gelman and coauthors found a similar reversal between states and individuals: richer states vote more Democratic while richer individuals within them vote more Republican [5]. Chapter 3.3, The income ladder follows the individual vote by family income.

What these data cannot show

The ZIP-code results are associations between places and do not say how a capital-gains filer or an EITC claimant voted. Group and individual correlations can differ [6], and this chapter has one case. The raw ZIP-code coefficient of the swing on the capital-gains share is +2.4, while among voters without a degree investors moved 5.5 points toward the Republicans and non-investors 3.5 points toward the Democrats. With the college share held fixed the ZIP-code coefficient is −1.0, the same sign as the survey difference, though its interval includes zero when clustered on state. The place-level and voter-level results disagree about the raw association and agree in sign once education is held fixed, which is what the composition account above predicts. This is ecological inference, and the ZIP-code filers and voters are overlapping but not identical populations.

The timing is a limit. The IRS shares are for tax year 2022 and the swing is for 2016 to 2020, so a ZIP code's income mix is measured after both elections. The release of the American Community Survey behind the ZIP-code controls is not recorded in the demographic file, so the controls cannot be dated against the swing. The IRS shares describe filers, so the poorest households and some retirees are missing, and the EITC share is a share of returns and not of residents. The vote estimates are allocations of precinct returns to ZIP codes, with a source change between 2016 and 2020, and both the allocation error and the source change enter the swing. The IRS file also omits PO-box ZIP codes and ZIP codes below its disclosure thresholds, so some ZIP codes with votes have no share; Figure 8 counts what each step of the sample leaves out.

Figure 8

The ZIP-code sample is 27,144 areas in 3,088 counties and the county sample is 3,080 counties

Counts of ZIP code tabulation areas and counties at each step from the vote file to the two analysis samples

Sample	ZIP code tabulation areas	Counties containing these areas
ZIP code tabulation areas with vote estimates for 2016 and 2020	32,428	3,102
With more than 100 votes in 2020	29,420	3,100
With both IRS shares and a swing	27,170	3,089
With the average tax rate and four ACS controls: the ZIP-code sample	27,144	3,088
Counties with a two-party return in 2016 and 2020	n/a	3,113
With ZIP-code shares	n/a	3,088
With county demographics: the county sample	n/a	3,080

Show data

Data for The ZIP-code sample is 27,144 areas in 3,088 counties and the county sample is 3,080 counties
counties	step	zcta
3,102	ZIP code tabulation areas with vote estimates for 2016 and 2020	32428
3,100	With more than 100 votes in 2020	29420
3,089	With both IRS shares and a swing	27170
3,088	With the average tax rate and four ACS controls: the ZIP-code sample	27144
3,113	Counties with a two-party return in 2016 and 2020	n/a
3,088	With ZIP-code shares	n/a
3,080	With county demographics: the county sample	n/a

The ZIP-code sample holds 27,144 of the 32,428 areas in the vote file, with 156.4 million of its 157.4 million votes in 2020; the county sample holds 3,080 of the 3,113 counties with returns. Each row keeps the areas of the row above that meet one more condition, and areas with 100 or fewer votes in 2020 are dropped at the second row. A row with no ZIP-code count applies only to counties.Note: For the ZIP-code rows the last column counts the counties that contain the areas. Of the counties with returns that have no ZIP-code shares, the Hawaii counties are not covered by the vote file, Kansas City, Missouri, is coded as a county of its own in the returns file, and small counties and Virginia independent cities lose their ZIP codes at the vote or IRS step. The Connecticut counties have ZIP-code shares and no county demographics in the panel and drop out of the county sample. A retired county code in the vote file (Shannon County, South Dakota, now Oglala Lakota County) is matched to its current code when the county sample is built.Source: Precinct votes allocated to ZIP code tabulation areas (Fekrazad 2025), 2016 and 2020; certified county returns, 2016 and 2020; full counts, no weights

The intervals allow dependence among ZIP codes in the same county (or, in the state check, the same state) and not between neighboring counties in different states or between neighboring states. Regional swings, such as the swing in the Texas border counties, make that dependence likely, and the two capital-gains intervals that change between the county and state clusterings show that it can matter for a conclusion. The shares of the smallest ZIP codes rest on few returns. In the sample, 1,405 ZIP codes report no capital-gains returns and 1,244 report no EITC returns, and the data do not distinguish a true zero from a small count rounded down. They hold 0.3% and 0.2% of the votes in the sample, and leaving them out of the lowest decile moves its mean swing from 7.2 to 7.5 points toward the Republicans for capital gains and from 6.3 to 6.4 points toward the Democrats for the EITC.

The capital-gains result cannot be resolved by more controls. The share correlates +0.81 with the college share, the ZIP-code college variable has no resolved definition, and the county results with the bachelor's share carry intervals that include zero, so the data cannot say how much of the capital-gains association is capital gains and how much is education. Every ZIP-code coefficient with the college share held fixed, including the negative capital-gains coefficient and the EITC coefficient near zero, is an adjustment for a variable whose coding is unresolved: if it counts some college or more, the adjustment removes a broader educational gradient than a four-year degree would. The results also depend on the weights. Figure 9 gives the partial correlation of the swing with each share under vote weights and with each ZIP code counted equally. The signs of the capital-gains partial correlation with the college share held fixed and of the EITC partial correlation with the full set held fixed differ between the weightings; the capital-gains partial correlation with the full set held fixed is negative under both.

Figure 9

Two of the partial correlations change sign when every ZIP code counts equally

Partial correlation of the 2016 to 2020 swing with each share, with vote weights and with every ZIP code counted equally; ZIP-code sample

Share	Controls	Partial correlation, vote weights	Partial correlation, each ZIP code equal
Capital-gains share	No controls	+0.33	+0.35
Capital-gains share	College share held fixed	−0.09	+0.04
Capital-gains share	Poverty, college, Black and Hispanic shares held fixed	−0.26	−0.05
EITC share	No controls	−0.47	−0.38
EITC share	College share held fixed	−0.22	−0.18
EITC share	Poverty, college, Black and Hispanic shares held fixed	+0.02	−0.04

Show data

Data for Two of the partial correlations change sign when every ZIP code counts equally
Share	Controls	Vote weights	Lower bound	Upper bound	Each ZIP code equal	Lower bound	Upper bound	ZIP codes
Capital-gains share	No controls	+0.330	+0.288	+0.376	+0.345	+0.318	+0.374	27,144
Capital-gains share	College share held fixed	−0.093	−0.145	−0.038	+0.037	+0.010	+0.063	27,144
Capital-gains share	Poverty, college, Black and Hispanic shares held fixed	−0.258	−0.296	−0.214	−0.051	−0.078	−0.021	27,144
EITC share	No controls	−0.475	−0.520	−0.431	−0.381	−0.410	−0.353	27,144
EITC share	College share held fixed	−0.222	−0.289	−0.158	−0.179	−0.215	−0.143	27,144
EITC share	Poverty, college, Black and Hispanic shares held fixed	+0.021	−0.051	+0.085	−0.042	−0.067	−0.017	27,144

The capital-gains partial correlation with the college share held fixed is −0.09 with vote weights and +0.04 with equal weights, and the EITC partial correlation with the full set held fixed is +0.02 and −0.04. A partial correlation is the correlation left after the controls are removed from both variables. The capital-gains partial correlation with the full set held fixed is negative under both weightings.Note: Each 95% interval is a percentile interval from a seeded bootstrap of counties and is in the data table. Every row uses the same ZIP codes.Source: Precinct votes allocated to ZIP code tabulation areas (Fekrazad 2025), 2016 and 2020; IRS Statistics of Income ZIP-code data, tax year 2022; ACS 5-year estimates; weighted by 2020 votes and, in the second column, unweighted

The wealth measures are narrow. The survey has stock ownership and home ownership as yes or no items, so neither measures the size of holdings. The vote in the survey is self-reported, and the intervals treat the weighted sample as a simple random sample. Chapter 13.1, Measures of hardship compared puts hardship measures on one specification.

Data sources

Precinct votes on census geographiesPrecinct votes allocated to ZIP code tabulation areas, contiguous states, 2016 and 2020 (Fekrazad 2025, RLCR), joined to the project's ZIP-code demographic file (ACS-based; release not recorded)Years: 2016 and 2020 (contiguous United States only) · Weights: 2020 votes in each ZIP code (pres_tot_2020); ZIP codes with 100 or fewer votes dropped · License: CC0 (data doi:10.7910/DVN/Z8TSH3), as stated in the project README.
IRS SOI ZIP code dataIRS Statistics of Income ZIP-code data, tax year 2022: share of returns claiming the earned income tax credit, share reporting a net capital gain, mean effective tax rateYears: Tax year as distributed with the ZCTA atlas file. · Weights: none (administrative counts by ZIP code); ZIP codes weighted by 2020 votes in the analysis · License: See the provider's terms of use.
County election returnsCounty presidential returns, 2016 and 2020, repaired file; the county swing and the county weightsYears: President 1868-2024 (40 elections); House and Senate 1990-2024 · Weights: none (full counts); county estimates weighted by 2020 two-party votes · 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, 2020 election-year row: bachelor's share, poverty, Black and Hispanic sharesYears: 2000-2024 (election years, interpolated between census and survey anchors) · Weights: none (full counts and interpolated estimates) · License: US government source data, public domain. The interpolated panel is derived work.
Cooperative Election StudyCooperative Election Study, 2016 and 2020: investor and homeowner items and college degree, respondents who voted for one of the two major partiesYears: 2006-2024 (every year; self-reported turnout is available in the 10 even years) · Weights: post-election weight (weight_post) · License: See the provider's terms of use.

All data sources

Methods

Sample. ZIP code tabulation areas (ZCTAs) in the contiguous states with vote estimates for 2016 and 2020, kept when they have more than 100 votes in 2020, both IRS shares, a swing, the average tax rate and the four ACS controls; this common sample of 27,144 ZCTAs in 3,088 counties is used for every ZIP-code estimate. County sample: 3,113 counties with returns in 2016 and 2020 (Alaska excluded), of which 3,080 have ZIP-code shares and county demographics. The sample table in the last section gives the counts at each step. Votes and outcome. Precinct returns allocated to ZCTAs by the RLCR method (2016 from VEST, 2020 from MEDSL), fractional counts; the allocation reproduces the national two-party Democratic share of 51.09% in 2016 and 52.24% in 2020. The outcome is the swing, the 2020 two-party margin minus the 2016 margin, Democratic minus Republican, in points, so a positive swing is toward the Democrats. The county swing comes from certified county returns. Measures. The share of returns claiming the earned income tax credit and the share reporting a net capital gain (IRS Statistics of Income ZIP-code data, tax year 2022, a year recorded in the project's build code and not in the data file) and the mean effective tax rate; the poverty rate, Black share, Hispanic share and college share from the ACS 5-year estimates in the project's ZIP-code demographic file, dated March 2025, which records neither the ACS release nor the table behind each variable, so these controls are undated. The county controls are the 2020 row of the county demographics panel. The definition of the ZIP-code college variable is unresolved. The file describes it as the percentage of the population with a college degree or higher and names no table, age group or release, but aggregated to counties it averages 52.7% against 23.6% for the county bachelor's share (correlation +0.86), so its level is far above a bachelor's share. The level and the correlation do not establish what the variable counts, so the chapter calls it the college share and leaves its coding open. Weights. ZIP-code estimates are weighted by 2020 votes in the ZCTA and county estimates by 2020 two-party votes. CES estimates use the post-election weight. Correlations and partial correlations are also computed with every ZIP code (or county) counted equally. Estimator. For each share, a weighted least squares regression of the swing on the share and the named controls, every variable standardized on the sample with the same weights, so the coefficient is points of swing per standard deviation of the share. Each set of controls is a separate model, and the full set (poverty, college, Black and Hispanic shares) does not include the average tax rate. Intervals and clustering. 95% intervals use cluster-robust (CR1) standard errors clustered on county for ZIP-code models and on state for county models (t distribution with the number of clusters minus one degrees of freedom). As a sensitivity check the ZIP-code ladder and the decile means are also computed with clusters on state (49 clusters: the states and the District of Columbia); this moves the two capital-gains intervals that hold the college share fixed, with and without the tax rate, to include zero and leaves the other ladder intervals on the same side of zero. Neither clustering allows for dependence between neighboring states. Correlations use a seeded bootstrap of whole clusters (2,000 replicates, seed 12345); decile means use the clustered ratio estimator; CES margins use the Kish effective sample size and a normal interval, which is model based and treats the weighted sample as a simple random sample. Suppression. ZCTAs with 100 or fewer votes are dropped; no survey cell rests on fewer than 30 respondents. Decile groups hold equal numbers of ZCTAs. The 2016 and 2020 survey samples are separate cross-sections, and voters who report neither owning nor renting are left out of the homeowner comparison.

References

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
Voting and Election Science Team (2018). 2016 Precinct-Level Election Results. Harvard Dataverse. doi:10.7910/DVN/NH5S2I
Baltz, Samuel; Agadjanian, Alexander; Chin, Declan; Curiel, John; DeLuca, Kevin; Dunham, James; Miranda, Jennifer; Phillips, Connor Halloran; Uhlman, Annabel; Wimpy, Cameron; Zárate, Marcos; Stewart III, Charles (2022). American election results at the precinct level. Scientific Data 9: 651. doi:10.1038/s41597-022-01745-0
Gethin, Amory; Martínez-Toledano, Clara; Piketty, Thomas (2022). Brahmin Left Versus Merchant Right: Changing Political Cleavages in 21 Western Democracies, 1948-2020. Quarterly Journal of Economics 137(1): 1-48. doi:10.1093/qje/qjab036
Gelman, Andrew; Park, David; Shor, Boris; Bafumi, Joseph; Cortina, Jeronimo (2008). Red State, Blue State, Rich State, Poor State: Why Americans Vote the Way They Do. Princeton University Press.
Robinson, W. S. (1950). Ecological Correlations and the Behavior of Individuals. American Sociological Review 15(3): 351-357.

How to cite

@misc{singh2026wagescapitalandthecountyvote,
  author = {Singh, Vishal},
  title = {Wages, capital and the ZIP code vote},
  year = {2026},
  url = {https://vishalsingh.org/politics/wages-capital-and-the-county-vote},
  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.

Income
- Chapter 3.3, The income ladder Individuals
- Chapter 12.3, Income and the county vote (planned) Places
- Chapter 8.7, System justification (planned) Countries

Reading paths

Income and class