Game Theory
The Public Goods Game: Why People Stop Contributing
The Public Goods Game is the laboratory workhorse of collective-action theory: a group of players each receive an endowment and secretly choose how much to put into a shared pot, which is then multiplied and split equally among everyone. The multiplier makes cooperation collectively optimal — every euro contributed grows the pie — yet each individual's private best move is to contribute nothing and free-ride on others. The dominant-strategy prediction is bleak: zero contribution. What makes the game famous is not that theory says zero, but that real people don't start at zero — they begin generously and then, watching free-riders out-earn them, unravel toward it. It is the cleanest experimental proof that individual rationality and group welfare can collide, and the standard model for taxes, climate treaties, open-source code, and Wikipedia.
- FieldExperimental & behavioral game theory
- Key parameterMPCR (marginal per-capita return), 0 < r/n < 1 < r
- Nash predictionContribute 0 (free-ride)
- Social optimumContribute everything
- Famous experimentFehr & Gächter, Nature 2000 — punishment sustains cooperation
- Real-world analoguesTaxes, climate treaties, open-source, Wikipedia, the commons
Interactive visualization
Press play, or step through manually. The visualization is yours to drive — try it before reading on.
Watch the 60-second explainer
A condensed visual walkthrough — narrated, captioned, under a minute.
The model, precisely
There are n players. Each gets an endowment y of tokens. Player i secretly contributes gᵢ (0 ≤ gᵢ ≤ y) to a common pot and keeps the rest. The pot is multiplied by a factor r (with 1 < r < n) and split equally among all n players. Player i's payoff is:
πᵢ = (y − gᵢ) + (r/n) · ∑ⱼ gⱼ
The private half-you-keep, plus your equal share of the multiplied common pot. The crucial quantity is the marginal per-capita return, MPCR = r/n. Because r < n, we have MPCR < 1: for every token you put in, you personally get only r/n < 1 back — a private loss. So ∂πᵢ/∂gᵢ = −1 + r/n < 0, and the selfish best response is gᵢ = 0 regardless of what anyone else does. That makes zero contribution a dominant strategy and the unique Nash equilibrium.
Yet r > 1, so each token creates r tokens of total value: the group as a whole earns r/n × n = r > 1 per token contributed. Full contribution maximizes the pie. The gap between the individual condition (r/n < 1) and the social condition (r > 1) — which requires only 1 < r < n — is the social dilemma. It is a continuous, n-person Prisoner's Dilemma.
A worked round of four players
Take the canonical lab setup: n = 4, endowment y = 20 tokens, multiplier r = 1.6, so MPCR = 1.6/4 = 0.4.
- Everyone contributes fully (all give 20): pot = 80, multiplied = 128, split = 32 each. Everyone turns 20 into 32 — a 60% gain. This is the social optimum.
- You free-ride while the other three give 20: pot = 60, ×1.6 = 96, split = 24. You keep your 20 and collect 24 = 44. The three cooperators get only 24. You out-earn them by 20 by contributing nothing.
- Everyone free-rides (all give 0): pot = 0, everyone just keeps 20. The public good never exists.
Notice the trap: 32 > 20, so cooperation beats universal defection — yet 44 > 32, so for any given behavior of others, defecting beats cooperating for you. Both statements are true at once. The 44-vs-24 spread is what a cooperator sees when a free-rider sits in their group, and it is exactly the signal that drives the decay animated in repeated play.
Why contributions collapse over repeated rounds
Run the game for, say, 10 rounds with the same or reshuffled groups and a striking regularity appears (Isaac & Walker 1988; hundreds of replications since). Round 1 contributions average roughly 40–60% of the endowment — far above the zero the theory predicts. Then they fall, round after round, ending near 10% or below by the final period. The public good visibly erodes.
The dominant explanation is conditional cooperation (Fischbacher, Gächter & Fehr 2001): most people are not pure free-riders and not pure altruists — they are reciprocators who match what they expect others to give. Fischbacher's data classify about 50% of subjects as conditional cooperators, ~30% as free-riders, and the rest as mixed or hump-shaped. This composition is combustible. Conditional cooperators start hopeful and contribute; the free-riders drag the group average down; the conditional cooperators observe the shortfall, revise expectations downward, and cut their own giving to match — which lowers the average again next round. The result is a self-reinforcing downward spiral, not a moral failing but a coordination failure among people who would cooperate if they trusted others to. The end-of-experiment 'restart effect' — reset the game and contributions jump back up — confirms it's disappointed expectations, not preference change, driving the collapse.
What stops the collapse: punishment, and the Fehr–Gächter result
The most cited public-goods finding is Ernst Fehr & Simon Gächter, “Cooperation and Punishment in Public Goods Experiments” (American Economic Review, 2000; extended in Nature, 2002). They added a second stage: after seeing contributions, players could pay to punish free-riders — spend 1 token to destroy 3 of a target's tokens. Standard theory says no one should ever punish (it's costly and helps others), so the threat is empty and behavior shouldn't change.
It changed dramatically. In the no-punishment treatment contributions decayed toward zero as usual. In the punishment treatment, subjects did pay to punish free-riders — this is altruistic (or 'strong reciprocity') punishment — and contributions rose over rounds, stabilizing near full cooperation (often 80–100% of endowment). The mere expectation of being punished disciplined would-be free-riders. Later work (Gürerk, Irlenbusch & Rockenbach, Science 2006) let subjects choose between a punishment world and a punishment-free world: people migrated toward the sanctioning institution and prospered. The lesson: decentralized peer enforcement can solve the dilemma the pure incentives cannot — but 'antisocial punishment' (spiteful punishment of high contributors, common in low-trust societies per Herrmann, Thöni & Gächter, Science 2008) can wreck it, which is why formal institutions and law exist.
Where this shows up in the real world
The public-goods game is not a toy — it is the abstract skeleton of most collective-action problems:
- Taxes. National defense, roads, and courts are non-excludable public goods; everyone benefits whether or not they pay. Voluntary funding under-provides them, which is the economic justification for compulsory taxation with penalties — the state as the mandatory 'punishment' stage.
- Climate treaties. Each nation prefers others to cut emissions (a global public good) while it free-rides. The Kyoto Protocol (1997) and Paris Agreement (2015) are attempts to build the enforcement stage — pledges, review, and reputational sanction — that voluntary contribution alone lacks.
- Open collaboration. Open-source software, Wikipedia, and scientific replication are public goods maintained by a minority of contributors while most 'free-ride' as users. Elinor Ostrom's Nobel-winning fieldwork (1990) on fisheries, irrigation, and forests showed real communities escape the tragedy of the commons using exactly the ingredients the lab identifies: monitoring, graduated sanctions, and communication.
- Communication itself. Simply letting players talk before contributing — 'cheap talk' with no binding force — sharply raises cooperation, one of the most robust results in the literature and a rebuke to the idea that only material incentives matter.
See Public Goods for the underlying non-rival, non-excludable property, and the Coase Theorem for the private-bargaining alternative to state provision.
Assumptions, critiques, and a common misconception
The key condition is simply 1/n < MPCR < 1: private incentive to defect, social incentive to contribute. Change the numbers and the dilemma dissolves. If MPCR ≥ 1 (r ≥ n), contributing pays you directly and everyone gives fully — no dilemma. As n grows with r fixed, MPCR shrinks and free-riding gets more tempting, matching the intuition that large groups under-provide public goods more severely (Olson, The Logic of Collective Action, 1965).
Critiques. (1) The standard model assumes players maximize only their own money; the persistent above-zero giving falsifies that narrow assumption and motivated models of social preferences (Fehr–Schmidt inequity aversion, 1999). (2) Results are sensitive to design: whether groups are fixed ('partners') or reshuffled ('strangers'), whether the horizon is known, whether subjects are students (WEIRD samples). (3) The steep end-game drop shows people do reason strategically about backward induction, complicating the 'people are just nice' story.
The misconception to avoid: the decline toward zero is not evidence that 'people are selfish' or that 'cooperation is naive.' The opposite. If everyone were a rational free-rider, contributions would start at zero and stay there — flat. The fact that they start high and decline is the fingerprint of conditional cooperators being let down by a free-riding minority. People aren't giving up on cooperation because they never wanted it; they're withdrawing it because they can't sustain it unilaterally without enforcement. The problem is institutional, not moral.
| Feature | Public Goods Game | Prisoner's Dilemma | Ultimatum Game |
|---|---|---|---|
| Number of players | n ≥ 2 (typically 4) | Exactly 2 | Exactly 2 |
| Choice | Continuous (how much to give) | Binary (cooperate / defect) | Offer split, then accept/reject |
| Nash equilibrium | Contribute 0 (all free-ride) | Both defect | Offer ≈0, accept anything >0 |
| Social optimum | Contribute everything | Both cooperate | Any full split of the pie |
| What real people do | Start ~50%, decay toward 0 | ~50% cooperate one-shot | Offer ~40%, reject <20% |
| Core force studied | Free-riding & conditional cooperation | Mutual defection | Fairness & spite |
Frequently asked questions
What is the Nash equilibrium of the public goods game, and why isn't it the social optimum?
Because each token you contribute returns only MPCR = r/n < 1 to you personally, your money-maximizing best response is to contribute nothing, no matter what others do — so 'everyone contributes 0' is the unique Nash equilibrium. But since r > 1, each token creates r > 1 tokens of total value, so the social optimum is for everyone to contribute fully. The equilibrium is a dominant-strategy outcome; the optimum is not individually stable. That gap is the entire point of the game.
What does MPCR mean and why does it matter?
MPCR is the marginal per-capita return, r/n — the amount you personally get back per token you contribute. If MPCR < 1, contributing is a private loss and free-riding dominates. If MPCR ≥ 1, contributing pays you directly and the dilemma vanishes. Raising MPCR (a bigger multiplier or a smaller group) reliably increases contributions in experiments, which is why small groups police public goods better than large anonymous ones.
If free-riding is the rational strategy, why do real people contribute at all?
Most subjects are 'conditional cooperators' who match what they expect others to give, plus a minority of altruists — not pure self-interested maximizers. They start with optimistic expectations and contribute 40–60% in round one. Contributions then decline because free-riders undercut those expectations, not because people were selfish from the start. Communication, reputation, and the ability to punish free-riders can reverse the decline entirely.
How does punishment fix the game?
Fehr & Gächter (2000, 2002) let players pay to reduce free-riders' earnings (e.g., spend 1 to destroy 3). Theory says no one should punish, yet people do — 'altruistic punishment' — and the threat pushes contributions up toward the full-cooperation optimum instead of down toward zero. It shows decentralized peer enforcement can solve the dilemma. The caveat is 'antisocial punishment' of cooperators, which can destroy cooperation in low-trust settings.
How is the public goods game different from the prisoner's dilemma?
It's essentially an n-person prisoner's dilemma with a continuous choice. The two-player prisoner's dilemma forces a binary cooperate/defect; the public goods game has many players choosing how much (0 to y) to give. Both share the structure that individual rationality (defect / give nothing) produces a group-worse outcome than mutual cooperation. The public goods game is preferred for studying groups, gradual free-riding, and the dynamics of decay over repeated rounds.
Is the collapse of contributions inevitable?
No. It's the default under anonymous, one-shot-style voluntary contribution, but Elinor Ostrom's fieldwork and later experiments show it's routinely avoided with the right institutions: monitoring, communication, reputation, graduated sanctions, and repeated interaction among a stable group. Even non-binding 'cheap talk' before contributing sharply raises cooperation. The collapse is a symptom of missing enforcement, not an iron law of human nature.