Skip to main content
Convention

When we say a candidate won by 5 points, we mean 5% of all votes cast.

Most political data sites report margin as two-party share — excluding third parties from the denominator. Our convention includes every vote. In most races the difference is tiny. When it matters, it really matters.
Back to data sourcesCanvas uses this too
Our convention
(D-R)/total
Share of all votes cast
Two-party share
(D-R)/(D+R)
The alternative most other sites use
Signal
>1 pt gap
Where the method starts to matter
Why this matters

The denominator changes the story

In a two-way race with almost no third-party votes, both conventions agree. But elections with significant third-party candidates — Perot in 1992, McMullin in 2016 — produce noticeably different margins depending on the method.
Our convention is inclusive
Every voter who cast a ballot is in the denominator. A candidate who wins 45% in a three-way race with 15% third-party support gets credit for being at 45 — not at a normalized 51.
Two-party share is comparative
Two-party share normalizes races against each other by stripping third parties from the denominator. Useful for comparing competitive margins across races with different third-party levels, but loses the absolute signal.
Try it

Calculate a margin both ways

Enter your own vote totals — or load a preset from a real election.
Interactive calculator
See the difference for any set of vote totals
Enter vote counts — or load a preset — and compare both methods side by side.
Load a preset
Akashic margin
(D - R) / total votes × 100
+1.23pts
Two-party share
(D - R) / (D + R) × 100
+1.23pts
Difference between methods
Small enough that most users will not notice.
0.00pts
Total votes cast:4,484,902
akashic_margin = (dem - rep) / total × 100
two_party_share = (dem - rep) / (dem + rep) × 100
Worked examples

Real races where the choice of method shows

2016 Utah
Trump +18 (our) · Trump +26 (two-party)
Evan McMullin took 21.5% of the state. Under our convention, Trump's lead over Clinton is roughly 18 points. Two-party share — with McMullin stripped out — reports a 26-point gap. Same race, very different sounding margin.
1992 Presidential (national)
Clinton +5.6 (our) · Clinton +6.9 (two-party)
Ross Perot took 18.9% nationally. Clinton's margin over Bush is 5.6 points of all votes cast. Strip out Perot and it inflates to 6.9 points. The gap shrinks the more third-party support there is.
2020 GA Senate Runoff
Ossoff +1.2 (both)
A clean two-way race with essentially no third party. Both conventions agree to the second decimal. Most contemporary competitive races look like this.
Three ways to quote a margin

Comparison at a glance

ConventionFormulaBest for
Akashic margin(D − R) / total × 100Absolute performance. Works identically in two-way and multi-way races.
Two-party share(D − R) / (D + R) × 100Comparing races across cycles. Normalizes away third-party noise.
Vote percentageD / total · R / totalReporting raw shares without implying a head-to-head.
When to care

Rules of thumb for picking the right number

Third-party share above 2%
The two methods start to diverge. If you are comparing our numbers to someone else's, confirm the denominator before calling them mismatched.
Compare-to-history workflows
Consistency matters more than which convention you pick. Stick to one method across all the races you are comparing.
Quoting in print
Use our convention for absolute performance. Call it 'share of all votes cast' to avoid confusion with two-party share.
Modeling competitive margins
Two-party share is cleaner for some regressions. We expose both when you query the API.