How it works, precisely
The voter grades EACH option independently of the others: they evaluate, they do not rank. Two options can receive the same grade.
For each option the grades are sorted and the median is taken — the grade such that a majority judges it « at least this good » and a majority « at most this good ». Unlike the mean, the median is insensitive to extreme grades cast for tactical reasons.
When medians tie, the tie is broken by removing median grades one by one from the tied options until they diverge. That amounts to comparing the proportions of supporters and opponents around the median grade.
A worked example
Three options graded by eleven people on a five-grade scale.
| Option | Grades received | Median |
|---|---|---|
| Bordeaux | 3 Very good, 5 Good, 3 Fair | Good |
| Lyon | 5 Very good, 1 Good, 5 Reject | Good |
| Lille | 2 Very good, 4 Fair, 5 Poor | Fair |
- Bordeaux and Lyon share the same median grade: Good.
- Remove one median grade from each and repeat: Bordeaux keeps a broader base above Good.
- Lyon, deeply divisive (5 « Very good » but 5 « Reject »), drops behind.
Majority judgment prefers the widely respected option to the polarising one. A single-round vote would have crowned Lyon on its five enthusiasts.
Where it comes from
Majority judgment was proposed in 2007 by Michel Balinski and Rida Laraki, researchers at École polytechnique and the CNRS, in the Proceedings of the National Academy of Sciences. Their reference book, « Majority Judgment: Measuring, Ranking, and Electing », was published by MIT Press in 2011.
The inspiration comes from outside politics: wine competitions, gymnastics and figure skating have long graded on verbal scales and discarded extremes, because averages are far too easy for a single judge to manipulate.
The authors tested it in April 2007 with voters in Orsay during the French presidential election, alongside the official ballot. Several French citizen primaries have used it since, including the Primaire populaire in 2022.
Where it is used
- Internal elections and citizen primaries, where you want to measure real support.
- Assessing applications, projects or suppliers on qualitative criteria.
- Decisions where you want to spot the divisive option before adopting it.
- Any vote where the intensity of support, not just its existence, should weigh.
Limits and pitfalls
- Expresses nuance, not just a choice
- Limits tactical voting
- Measures real support
- The median is not intuitive
- A longer ballot
- Lots of possible ties
Frequently asked questions
Why the median rather than the mean?
Because the mean can be bought. A tactical voter giving the minimum grade to a serious rival shifts their mean considerably; it shifts the median by one notch at most, and only if many do the same. The median is the heart of the method, not a computational detail.
Is it genuinely unmanipulable?
No — the Gibbard-Satterthwaite theorem forbids that for any method. But majority judgment sharply limits the expected gain from an insincere vote, which is the best that can be proved.
How many grades should there be?
Five or six, with clear verbal labels. Fewer and you lose the nuance; more and voters stop distinguishing neighbouring steps.