Where Electoral Votes Are, and Where They Are Contested

The test

The vote columns come from Jason Timm’s PresElectionResults compilation (https://github.com/jaytimm/PresElectionResults), state-level shares for the 2024 presidential election, resting on the fifty-one certified state canvasses (the checked and signed official totals). The electoral votes are the 2020 census apportionment, fixed in law until 2032.² This resolution is the right one, not a simplification of something better: the rule under examination operates on whole states, so a 51-row table is the object the law actually acts on.

No federal body adds up American election returns, so the compilation is a private assembly of fifty-one official publications — the ordinary way election data reaches anyone. What a certified return is, and why the national file is always somebody’s private act, is the returns-source chapter’s subject.

The data

One row per jurisdiction: fifty states and the District of Columbia, which has had three electoral votes since the Twenty-Third Amendment was ratified in 1961.³ Here is Pennsylvania’s.

That row is three different kinds of thing sitting in a single line. The percentages, harris, trump, other and the computed margin, were measured: counted by county officials, certified by a state board, assembled by the compiler. The ev column was legislated: it was entered by hand from the apportionment, fifty-one numbers checked once against their required sum of 538. It is nonetheless the most certain number in the file, because it is a legal fact rather than an estimate. And two columns, col and row, were drawn: they are a design decision placing each state on the tile grid of the figures below, so the maps can show electoral weight without showing land. Nothing in the row records which is which.

Two cleaning decisions set this brief’s limits. First, the compilation records a third-party figure only sporadically, and the build turns every missing other into a zero. 42 of 51 jurisdictions therefore report a zero that is not a zero, and not one state’s three percentages add to 100. States reporting a zero fall short of 100 by 2.06 points on average, against 1.34 where a number was recorded. The shortfall is the third-party vote, sitting there unlabeled. A zero and a missing value are not the same thing, and a spreadsheet cannot tell them apart.

Second, one row per state with one winner per state assumes winner-take-all, and two states do not use it. Maine and Nebraska award two electors statewide and one per congressional district,⁴ and there is no cell in this schema in which a split could be expressed. A schema is a claim about how the world is arranged, and this one quietly claims that winner-take-all is universal. The cost of that claim is measured below.

Before you read the findings, commit to a winner and to a margin. Under winner-take-all the 2024 result was 312–226, a margin of 86 electoral votes. Now give every state’s electors out in proportion to its two-party vote instead — its Harris and Trump votes only, third parties set aside. Does the winner change — and what is the new margin? Write down a name and a number. They are two separate questions, and one answer is much larger than most people expect while the other is not what the size of that change would suggest.

What it says about democracy

Start with the map, drawn the only honest way a state-level rule can be drawn: land does not vote, so no land is drawn.

Color is the margin, capped at plus or minus 30 points, so the 10 jurisdictions beyond the cap all read as the same extreme; the largest is District of Columbia at -83.8. Hover for detail, and use the buttons to give each square an area proportional to its electoral votes.

Figure 1. The 2024 result as a grid, drawn twice. One square per jurisdiction fits this data, because the rule acts on whole states. Each square sits roughly where the state belongs and is colored by margin, Trump’s percentage minus Harris’s, on a scale capped at 30 points in each direction. In the first version every square is the same size, so every state counts once; in the second, a cartogram — a map that sizes each place by a number rather than by its land — each square’s area is proportional to its electoral votes, which is what the Constitution actually counts. 10 of the 51 jurisdictions lie beyond the cap and are painted the same extreme shade; the largest is District of Columbia at -83.8.

Read the legend before you read the map. The margin runs from -83.8 in District of Columbia to 45.8 in Wyoming, a spread of 130 points, and the capped scale cannot show the difference among the blowouts. A cap makes the middle of the scale readable by making the ends of it lie. Every choropleth — a map that colours areas by a number — you will ever see has made this trade; most do not tell you.

The color itself is a choice too. Color the same squares by Harris’s share instead of the margin, changing nothing else, and the country pales:

Identical geography, identical data, identical color scale; only the quantity being colored has changed. The right-hand map is the paler one because a margin is about 2.0 times the distance of a share from 50. Hover either map.

Figure 2. Margin and share, side by side, on one color ramp. The same 51 squares, the same blue-white-red scale, the same cap of 30 points either side of a tie. On the left, color is the gap between the two candidates; on the right it is how far Harris’s own share sits from 50. 10 jurisdictions reach the cap on the left and 1 on the right, and nothing about the country differs between the panels.

The reason is arithmetic rather than politics. A margin counts the same voters twice, once as a gain for one candidate and once as a loss for the other, so it is about 2.0 times a share’s distance from 50. Neither map is lying. The choice of scale is an argument, and the only dishonest version is the one that does not tell you which was chosen.

What this brief cannot tell you

There are no people in this file. The vote columns are percentages and ev is a legal quantity, so turnout, population and votes per elector cannot be computed here, and the per-voter inequality the small-state argument turns on is out of reach. Everything is state-level, which hides the rule’s most important feature: a state’s losing voters vanish, and the file has no unit small enough to show the disappearance.

It cannot separate the rule from the country. Close states would be more interesting than safe ones under almost any system; winner-take-all sharpens the difference, and nothing here separates the sharpening from the underlying distribution.

It is one election. Competitiveness is a property of a state in a given election, not of the state: the Solid South was uncompetitive for eighty years and then was not, and Vermont ran the same path in reverse. The 8 competitive states of 2024 are a snapshot of a single November.

And the reform arithmetic holds the world constant. Change the rule and you change the campaign, the turnout and probably the vote itself. What proportional allocation would actually do is not what it does to a fixed set of votes.

What you should have learned

• 12 states hold a majority of the Electoral College, but only 4 of them were decided by under five points: where the electoral votes are and where they are contested are nearly disjoint questions.
• The 2024 election’s entire uncertainty sat in 8 states holding 17.8% of the College, and that list is a property of the election, not of the states.
• The Electoral College’s two inequalities are independent: proportional allocation would have cut the 2024 margin from 86 electoral votes to 10 without changing the winner, and without touching the two-senator bonus.
• A margin scale spans about 2.0 times what a share scale spans for the same result, because it counts every voter twice; which one a map uses is an argument.
• In this file, 42 of 51 jurisdictions report a third-party zero that is really a missing value, and no state’s percentages sum to 100.

Extensions

• pres2024_states.csv has ev and margin for every state. Sort by margin and add up the electoral votes of the closest states until you cross 270. How much of the map was actually in play?
• The other column of pres2024_states.csv is zero in most rows. Compute each state’s shortfall from 100 using harris, trump and other, and compare the states with a zero against the states without one. What is the zero hiding?
• Using ev and winner in pres2024_states.csv, find the smallest combination of states that reaches 270. How few jurisdictions is that?
• Stretch. The same compilation carries every election back to 1864. Download 2020, rebuild Figure 3, and see how much of the competitive list survives one cycle.
• Stretch. Maine and Nebraska publish district-level results. Download them and compute how often, since 2000, a split elector would have decided the presidency.

Sources

Data

• State-level 2024 presidential returns, from Jason Timm’s PresElectionResults compilation (https://github.com/jaytimm/PresElectionResults), which rests on the certified state canvasses. Electoral vote counts reflect the 2020 census apportionment.
• U.S. Census Bureau. 2020 Census Apportionment Results, the House seats, and so the electors, each state holds through 2028. https://www.census.gov/data/tables/2020/dec/2020-apportionment-data.html
• Maine and Nebraska’s district rules: 21-A M.R.S. §802 and Neb. Rev. Stat. §32-710. https://legislature.maine.gov/statutes/21-A/title21-Asec802.html https://nebraskalegislature.gov/laws/statutes.php?statute=32-710

The data itself

Every figure above is drawn from these tables. They are plain CSV, they open in a spreadsheet, and they are yours to take further.

pres2024_states.csv

Academic literature

• Edwards, G. C. III (2004). Why the Electoral College Is Bad for America. New Haven: Yale University Press. https://yalebooks.yale.edu/book/9780300109689/why-the-electoral-college-is-bad-for-america/
• Gelman, A., Katz, J. N. and Bafumi, J. (2004). “Standard Voting Power Indexes Do Not Work: An Empirical Analysis.” British Journal of Political Science 34(4): 657–674. https://doi.org/10.1017/S0007123404000237
• Chiafalo v. Washington, 591 U.S. 578 (2020), and McPherson v. Blacker, 146 U.S. 1 (1892). https://www.law.cornell.edu/supremecourt/text/19-465 https://www.law.cornell.edu/supremecourt/text/146/1

I want to rebuild a public dataset from its original source, without
any starting files. Please do the following and show your work.

THE SOURCE. Jason Timm's PresElectionResults collection on GitHub,
the state-level presidential returns file:
https://github.com/jaytimm/PresElectionResults/raw/master/data/pres_by_state.rda
No account or key is needed. It is an .rda file -- R's own data
format -- so read it in R, or with a library that can open R data
files. One row is one state in one election year; keep 2024. The
compilation is assembled from public election records; it is
somebody's careful stapling of fifty-one official sources, not an
official file itself, because no official national file exists.

One thing the source does not carry: electoral votes. Those are a
fact about the 2020 census apportionment, not about the returns, so
transcribe them yourself -- fifty-one small numbers -- and prove the
transcription by checking the sum.

WHAT TO BUILD.
1. One row per state plus the District of Columbia: the Harris,
   Trump, and other vote shares, the winner, the margin, and the
   state's electoral votes.

CHECK YOUR WORK. The copy this book used was downloaded in August 2026.
If your rebuild matches it, you should find:
- 51 rows, and electoral votes that sum to exactly 538
- Trump carries 31 of the 51, Harris 20
- the two biggest prizes are California at 54 and Texas at 40
The 2024 returns are settled history and should not move; the
collection itself grows by one election every four years.