Chapter 6.3

Precincts within counties

In 2024, 65.0% of two-party presidential votes in the sample were cast in precincts decided by 20 points or more, against 57.2% in counties decided by that margin.

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 64 Louisiana parishes are missing because the 2024 precinct file holds about half of their votes. Absentee, early-vote and provisional buckets are kept as reporting units and checked in a side calculation. The file has no precinct boundaries, so the chapter cannot map neighborhoods; that belongs to the planned Neighborhoods explorer.

Bucks County, north of Philadelphia, finished the 2024 presidential election at a two-party margin of −0.1 points, a Trump lead. Centre County, in the middle of Pennsylvania, finished at +2.9 points, a Harris lead. The margin is the Democratic vote minus the Republican vote over their sum, so a positive value is a Democratic lead, and both counties finished nearly even. Inside their borders the votes were distributed differently. In Bucks, many votes were cast in precincts close to the county's own nearly even result. In Centre, the votes piled up in precincts that one party or the other won by a wide margin, with few votes cast in precincts near even.

Precincts are more lopsided than the counties that hold them

A precinct here is a reporting unit, the area for which an election office reports its votes, and not a neighborhood. A county's margin is the average of its precinct margins weighted by their votes, so the color of a county on a map is a summary of the precincts inside it. The spread is what that summary leaves out. The spread is the standard deviation of a county's precinct margins around its own margin, weighted by each precinct's two-party votes and stated in the same points. It is 14.8 points in Bucks and 36.3 in Centre. Across the counties with enough precincts to measure one, with each county counting in proportion to its votes, the mean spread in 2024 is 23.3 points.

The same votes look different when counted by precinct and by county. A landslide here is a margin of 20 points or more, and a close contest is a margin under 10 points. In 2024, 65.0% of the two-party presidential votes in the sample were cast in landslide precincts, against 57.2% cast in landslide counties. Close contests run the other way, with fewer votes cast in close precincts than in close counties. The Pennsylvania counties show how one county result can hold either kind of precinct.

Figure 1

Bucks County precincts cluster near an even split, while Centre County precincts form two groups, one on each side

Share of each county's 2024 two-party votes cast in precincts in each 10-point band of margin

Share of the county's votes (%)

Show data

Data for Bucks County precincts cluster near an even split, while Centre County precincts form two groups, one on each side
county	n	share	x0	x1
Bucks County, PA	0	0	−100	−90
Bucks County, PA	0	0	−90	−80
Bucks County, PA	0	0	−80	−70
Bucks County, PA	0	0	−70	−60
Bucks County, PA	0	0	−60	−50
Bucks County, PA	0	0	−50	−40
Bucks County, PA	4	1.05	−40	−30
Bucks County, PA	25	8.35	−30	−20
Bucks County, PA	54	17.71	−20	−10
Bucks County, PA	83	27.62	−10	0
Bucks County, PA	68	22.48	0	10
Bucks County, PA	40	13.77	10	20
Bucks County, PA	15	5.06	20	30
Bucks County, PA	13	3.63	30	40
Bucks County, PA	2	0.315	40	50
Bucks County, PA	0	0	50	60
Bucks County, PA	0	0	60	70
Bucks County, PA	0	0	70	80
Bucks County, PA	0	0	80	90
Bucks County, PA	0	0	90	100
Centre County, PA	0	0	−100	−90
Centre County, PA	0	0	−90	−80
Centre County, PA	1	0.333	−80	−70
Centre County, PA	5	4.17	−70	−60
Centre County, PA	8	6.12	−60	−50
Centre County, PA	12	6.15	−50	−40
Centre County, PA	7	7.06	−40	−30
Centre County, PA	8	10.56	−30	−20
Centre County, PA	5	5.43	−20	−10
Centre County, PA	3	4.12	−10	0
Centre County, PA	2	1.34	0	10
Centre County, PA	7	11.96	10	20
Centre County, PA	8	14.10	20	30
Centre County, PA	7	12.99	30	40
Centre County, PA	7	9.71	40	50
Centre County, PA	3	2.83	50	60
Centre County, PA	4	3.13	60	70
Centre County, PA	0	0	70	80
Centre County, PA	0	0	80	90
Centre County, PA	0	0	90	100

Bucks's bars pile up on either side of the Even line, and Centre's form two groups, one on each side, with little between them. Each bar is the share of the county's 2024 two-party votes cast in precincts within one band of margin, with Democratic margins to the right. In Bucks, 50% of votes were cast in close precincts; in Centre, 5.5% were. Values in Table 1.

Note: The number in the tooltip is the number of precincts in the band. The two counties were chosen for the contrast between them, not at random.Note: Full counts of votes: there is no sampling, so the shares carry no confidence interval.Source: MIT Election Data and Science Lab, 2024 precinct-level presidential returns; county totals summed from the precincts; full counts of votes, not survey estimates, so nothing is weighted except by two-party votes where stated

Full counts of votes, in the units election offices report

These are counted votes, not survey estimates. The returns are the 2024 presidential results by precinct from the MIT Election Data and Science Lab, with the Harris and Trump votes kept and other candidates left out. A county enters the sample when its precinct rows add up to its certified total, which leaves 175,345 reporting units in 3,043 counties; Alaska and Louisiana are absent. A county needs at least 5 reporting units to have a spread. The qualification that matters is the unit. A precinct is whatever the state or county office reports, so precincts differ in size and kind from state to state, and some rows are not places at all. Rows labelled absentee, provisional, early voting, mail, failsafe, military or overseas, and rows labelled as statistical adjustments, are buckets of ballots, and they hold 1.5% of the votes in the sample. A full count has no sampling error, so shares and spreads carry no intervals, and a statistic computed across counties, such as a mean or a correlation, carries a 95% interval from resampling counties. The report gives every estimate with its interval and sample size.

Spread is widest in Mississippi and Alabama

Figure 2 maps the spread of every county that has one. The vote-weighted mean spread is 38.2 points in Mississippi, the highest of any state, with Alabama second. Among the counties with at least 100,000 two-party votes, the widest spreads are in Kings County, New York, and in Jefferson and Mobile counties in Alabama. The narrowest are in Arlington County, Virginia, and the District of Columbia.

The map carries a complication. The units that election offices report differ in size by state, and smaller units can show more spread, so a comparison of spread across states mixes the electorate with the way votes are counted. A county with few reporting units also has a spread that one unit can move a long way. The chapter's controls for a county's size and its state, described below, address part of that mix and not all of it.

Figure 2

Spread inside counties is widest in Mississippi and Alabama and scattered across the Southwest and Plains

Vote-weighted standard deviation of precinct margins inside each county, 2024, in points of two-party margin; counties with at least 5 reporting units

Loading county values

Spread of precinct margins inside the county (points)

Values outside 0 to 60 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.

Dark counties, where precinct margins vary most, cluster in Mississippi and Alabama, with scattered dark counties in the Southwest and the Plains. The shade is the spread of precinct margins inside the county, in points of two-party margin; counties with too few reporting units for a spread are drawn as no data. Switch to the margin view and the widest-spread counties are neither the reddest nor the bluest. Values in Table 3.

Note: Counties with fewer than 5 reporting units (190 of the 3,043 in the sample), the 64 Louisiana parishes, 6 other counties whose precinct rows do not match the certified total, Alaska and Kansas City, Missouri are drawn as no data.Note: A county near the floor of 5 reporting units has a spread that one unit can move a long way.Note: Full counts of votes: there is no sampling, so county margins and spreads carry no sampling error.Source: MIT Election Data and Science Lab, 2024 precinct-level presidential returns; county totals summed from the precincts; full counts of votes, not survey estimates, so nothing is weighted except by two-party votes where stated

Chance alone would leave precincts about as lopsided as counties

Part of the gap between precincts and counties could be a matter of size. Because each county margin is the average of its precinct margins, precinct margins are at least as spread out as county margins. That alone does not put more votes in landslide precincts, since a county decided by a landslide can split into precincts of which some are not. The simulation in Figure 3 is a benchmark for how much of the gap chance alone would produce. In each run, every precinct keeps its number of votes and its county's Democratic share, and its Democratic votes are drawn at random. It is as if the county's voters had been dealt to its precincts without regard to where they live. The simulated precincts hold 56.6% of votes in landslide precincts, close to the county share rather than the precinct share, and the same holds in the close band and the very lopsided band. Small units alone therefore do not produce the gap. Precinct margins inside a county vary more than random assignment of voters would make them.

Figure 3

Precincts are more often lopsided, and less often close, than counties or than random assignment of voters would make them

Share of 2024 two-party presidential votes cast in units decided by each margin, in percent of votes

Precincts
Counties
Simulated precincts

Show data

Data for Precincts are more often lopsided, and less often close, than counties or than random assignment of voters would make them
band	share	unit
Under 10 points	18.52	Precincts
Under 10 points	23.13	Counties
Under 10 points	23.32	Simulated precincts
20 points or more	64.97	Precincts
20 points or more	57.25	Counties
20 points or more	56.58	Simulated precincts
40 points or more	36.37	Precincts
40 points or more	25.38	Counties
40 points or more	24.96	Simulated precincts

In each band, compare the precinct bar with the other two: precincts hold more of the votes in landslides and fewer in close contests, and the simulated bar sits close to the county bar. The bands are close contests, landslides, and the very lopsided places decided by 40 points or more. The simulated bar is the mean over the random-assignment runs. Values in Table 2.

Note: Full counts of votes: there is no sampling, so the shares carry no confidence interval.Note: The simulated bars are means over 200 runs and move by at most 0.4 points across runs.Note: The sample is 175,345 reporting units in 3,043 counties: Alaska, the 64 Louisiana parishes and 6 other counties whose precinct rows do not match the certified total are left out.Source: MIT Election Data and Science Lab, 2024 precinct-level presidential returns; county totals summed from the precincts; full counts of votes, not survey estimates, so nothing is weighted except by two-party votes where stated

What spread goes with, and what it does not

Spread is higher in metro counties than in the least urban ones, and it is higher in counties that lean Democratic. Weighted by votes, the correlation of a county's margin with its spread is +0.4. Bluer counties are also larger, and larger counties have more precincts and more room for very different ones. With a county's votes and its number of precincts held fixed, and with counties compared only within their own state, the correlation falls and stays positive. Spread also goes with the nonwhite share of a county's residents, and that tie weakens less under the same controls than the tie to the rural share does. Spread is lower in counties farther from an even split, whichever party leads, and the chapter reports that pattern without reading it as a finding about competitiveness, because the data cannot say what produces it. Among counties with a margin of 50 points or less, the vote-weighted correlation has an interval that includes zero, and the negative link grows as the most lopsided counties are added.

Bucks and Centre are both among the counties the chapter calls near-tied, and that group differs widely in spread (Table 8). Bucks sits in its lowest bin and Centre past the line the chapter calls wide. None of this says why precinct margins vary more in some counties than in others. Spread goes with lean, size, nonwhite share and place type, and the controls are few and linear. A single spread also cannot tell a county with two humps, like Centre, from a county with a long tail.

What one election's precincts cannot tell you

The chapter describes 2024 alone and says nothing about whether precinct margins are spreading or narrowing; Planned, Sorting by geographic scale, 2016 to 2024 (planned), a planned chapter, will compare precincts across elections. The file has no precinct boundaries, so the data cannot say which precincts are adjacent, and a county with a wide spread need not have opposed precincts next to each other. A precinct margin describes a place and not the voters in it, and a landslide precinct still holds voters of the losing party. A pattern across places can differ from the pattern among the people who live in them, the problem of ecological inference [1]. Shares and spreads also change with the size and shape of the areas [2], and this chapter uses the units that election offices happened to report.

Go deeper

The report: every estimate with its interval and sample size.
Tables: every figure of the report as a table.
Elections, 1976 to 2024: the explorer for this part.

Reading paths that include this chapter:

Data sources

Precinct presidential returnsMIT Election Data and Science Lab, 2024 presidential returns by precinct (Harvard Dataverse), two-party votes per reporting unitYears: 2016, 2020, 2024 · Weights: none (full counts); statistics weighted by two-party votes · License: See the provider's terms of use.
County election returnsCertified 2024 county totals from the repaired county file, used to check each county's precinct rowsYears: President 1868-2024 (40 elections); House and Senate 1990-2024 · Weights: none (full counts) · 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 panel: ACS 2024 nonwhite share, rural share and Rural-Urban Continuum Code 2023Years: 2000-2024 (election years, interpolated between census and survey anchors) · Weights: none (county values; counties weighted by two-party votes in the correlations) · License: US government source data, public domain. The interpolated panel is derived work.

All data sources

References

Robinson, W. S. (1950). Ecological Correlations and the Behavior of Individuals. American Sociological Review 15(3): 351-357.
Openshaw, Stan (1984). The Modifiable Areal Unit Problem. Concepts and Techniques in Modern Geography 38, Norwich: Geo Books.

How to cite

@misc{singh2026precinctswithincounties,
  author = {Singh, Vishal},
  title = {Precincts within counties},
  year = {2026},
  url = {https://vishalsingh.org/politics/precincts-within-counties},
  note = {In The Public Mind. Published 30 September 2026.}
}