Developed by Lloyd S. Shapley (A Value for n-Person Games, 1953).
Divides the value a group creates by averaging each member's added contribution across every order in which the group could have formed.
Each partner's fair share is its added value, averaged over every order the team could form in
An engineering firm, a builder, and a financier bid together for a $12 million contract. Set what each could win alone and what each pair could win without the third. Each partner's share is the average of what it adds on joining, over the six orders the three could join in: bars are the shares, the black line is what each wins alone, and the shares always add up to the $12 million. Illustrative.
Fair share of the $12 million, $MWhat each wins alone, $MAn equal three-way split
An example
An engineering firm, a builder, and a financier won a $12 million contract to design, build, and fund a water treatment plant for a mid-sized city. Then they had to split the profit. The builder argued for an equal three-way split; the engineers argued for shares by hours worked; the financier, whose guarantee had made the bid credible, argued for more than either would give.
Their lawyer proposed a test all three could accept: estimate what each partner could have won alone and what each pair could have won without the third, then give each partner its added value averaged over every order in which the team might have come together. The estimates took an afternoon. Alone, the financier could win nothing, but its guarantee lifted every pair it joined. The builder could win small jobs alone and was the hardest to replace. The averages came to about $5.3 million for the builder, $4.3 million for the engineers, and $2.3 million for the financier, and each partner could check the arithmetic.
Why it works
When partners create value together, the obvious splits each favor someone. An equal split rewards the partner who adds least; a split by effort ignores the partner whose contribution is a guarantee or a license; a split by bargaining rewards whoever argues hardest. The Shapley value asks a narrower question: how much does each partner add, on average, when it joins?
Averaging over every joining order is the key move. A partner that joins a team already strong adds little; one that joins a weak team adds a lot. No order is privileged, so no partner gains from being first or last. The result has properties people find fair once they see them: the shares add up to the whole, a partner that adds nothing gets nothing, and two partners that add the same get the same.
The thinking behind it
Lloyd S. Shapley Lloyd S. Shapley, a game theorist and Nobel laureate, introduced his value in A Value for n-Person Games (1953), in the second volume of Contributions to the Theory of Games, edited by Harold W. Kuhn and Albert W. Tucker. The Shapley value divides the gains of a coalition by each player's average marginal contribution across all the orders in which the coalition could form. Each player's share depends only on what every group of players could achieve, so the method needs no view of who worked hardest or argued best. With three partners the method needs only what each partner and each pair could win, six estimates and one fixed total, which is all the figure asks for.
How to run it
- Define the partners and the prize. Name each member and the total value to be divided, after shared costs.
- Estimate what each subset could win. For three partners that is three solo values and three pairs; use bids lost, past jobs, or credible quotes.
- Agree the estimates before computing. The arithmetic is mechanical, so the arguments belong here, while no one yet knows their share.
- List every joining order. Three partners can join in six orders; four can join in twenty-four.
- Record each partner's added value in each order. Subtract the value of the team before it joined from the value after.
- Average each partner's additions. That average is the partner's fair share, and the shares always add up to the total.
- Check the result against what each wins alone. No partner should get less than it could win by itself; if one would, revisit the estimates.
Where it goes wrong
The common failure is arguing over the shares instead of the inputs. Once partners see the formula, each one argues for the subset estimates that favor it. Agree the estimates first, in writing, before anyone computes a share.
A second failure is too many partners. The number of orders grows fast (five partners have 120), and estimating every subset becomes guesswork. Group small contributors into one, or use sampling across orders. The third is assuming the shares are stable. When one partner could break away and do better with a single other partner, the split can be fair by this measure and still collapse. Check each pair's value against the shares its members get before signing.
When to use something else
- BATNA and ZOPA. When the split is being negotiated and the question is each side's walk-away point.
- Nash Equilibrium. When the partners act separately and the question is which choices are stable.
- Mechanism Design. When the task is designing the rules for sharing before the partners join, so each has a reason to report honestly.
Where it sits in SOLVE eX
Go to the source
- A Value for n-Person Games, Contributions to the Theory of Games, Volume II, Princeton University Press, 1953