Brief — patterns of collective action

Beads: course-wse.33 (P0). Blocks course-wse.23 (chapter 1 patterns module). Books to obtain: course-wse.44. Status: v4, signed off by Rufus 29 Oct ("go ahead on all fronts"). Pass 1 done and citation-checked; awaiting Rufus's decision below.

Question

What is the simple top-level typology of collective action problems that actually matters, and what lies beneath it?

The course uses three patterns now: the battle line (prisoner's dilemma), the stag hunt (assurance) and the volunteer (chicken). Rufus doesn't find the volunteer helpful. His challenge, in his words: "aren't many of these things the same thing? I don't get how a volunteer's dilemma or playing chicken is actually any different from the prisoner's dilemma."

The test of difference is structural, not a matter of framing. Write each pattern as a payoff matrix and ask what the equilibria are. In the stag hunt there are two equilibria; in the prisoner's dilemma there is only one. Patterns whose matrices have the same equilibrium structure are the same pattern under different names.

Pattern, as used here: a named strategic structure with a typical way of failing and typical fixes.

Shape of the answer

Top level: two axes, kept simple for the audience.

1. Equilibrium structure (from the payoff matrix). Candidate categories, to be tested:
   • One equilibrium, and it's bad: the prisoner's dilemma. An incentive problem.
   • Two equilibria, one better than the other: the stag hunt. A trust problem.
   • Two equilibria where someone has to give way: chicken, and the volunteer's dilemma. A who-pays problem. Test whether this is different enough from the prisoner's dilemma to matter.
   • Two equilibria, which convention to pick: pure coordination, and the battle of the sexes.
2. Taking or building. A negative commons that can be overexploited (fish, the atmosphere), or a positive commons that has to be built (information commons, open source, public goods). This is Ostrom's appropriation vs provision.

Second level: finer axes, useful but not the headline.

• How many players: two, or many (n-person).
• How contributions add up: sum, best shot (one contribution is enough), weakest link (everyone must contribute), threshold.
• Shape of the return to joining: Schelling's binary-choice curves, Heckathorn's payoff curves, and Marwell and Oliver's critical mass.
• Information, and one-shot vs repeated play.

Second layer: next to the patterns, not part of them.

• The principal-agent problem (moral hazard, adverse selection) is what a group faces once it hands a job to someone: "who monitors the monitor" (Alchian and Demsetz).
• Hold-up and lack of commitment are causes.
• Repeated play and the folk theorem, assurance contracts, and the Clarke–Groves and VCG mechanisms are remedies.
• Each item is placed as pattern, cause, remedy, or "adjacent, not a collective action problem".

Layers and passes

• Layer 1, structure. The games and patterns themselves. This is pass 1 and pass 2.
• Layer 2, evidence. The empirical and other literature: lab and field evidence on each pattern, how often each occurs, and what solves it. This is pass 3, which gets its own brief once layer 1 is settled.

Objectives and key results

Pass 1: structure and recommendation

O1. Settle the top level.

• KR1.1: Each candidate pattern written as a payoff matrix, with ordinal payoffs and its pure and mixed equilibria. Use the standard classification of 2×2 games (Rapoport and Guyer; Robinson and Goforth) to check that the categories are complete.
• KR1.2: A direct answer to "are chicken and the volunteer's dilemma the same as the prisoner's dilemma?", at the level of the matrix, and then at the level of what goes wrong and what fixes it.
• KR1.3: A recommended top level of at most four categories, plus the taking-or-building axis, with reasons it suits the audience.

O2. Map the typologies and finer axes.

• KR2.1: The typologies of Olson, Hardin, Ostrom, Kollock, Heckathorn and Sandler each summarised, with citation, pages and axes, and mapped onto the top level.
• KR2.2: The second-level axes listed, with which patterns they split or merge.
• KR2.3: Every claim about a source checked against the source. Anything not checked is marked [?].

O3. Place the second layer.

• KR3.1: Principal-agent (moral hazard, adverse selection) and at least four other mechanism-design or contracting items each placed as pattern, cause, remedy or adjacent, with a reason.

O4. Recommend, and hand Rufus a decision.

• KR4.1: A recommended pattern set for the course. Reasons are tied to the course/research/shortlist.md criteria: a case should teach a reusable mechanism rather than an outcome, and should overturn an assumption the audience already holds.
• KR4.2: Test: every case chapter 1 uses, and every case in course/research/shortlist.md, gets classified by the recommended set. Report cases no pattern fits, and cases that fit two patterns.
• KR4.3: A direct answer on the volunteer: keep, replace (with what), or demote to a footnote.
• KR4.4: A decision section Rufus can answer in one line, with what each answer means for course-wse.23.

Pass 2: knowledge base write-up

• KR5.1: A hub page, Patterns of collective action.md (tags: [moc]), with the top level, the finer axes and the typologies. Game theory basics.md links to it.

• KR5.2: An annotated bibliography page, Collective action bibliography.md, with one or two lines per source and a note on what it is good for. (Drafted.)

• KR5.3: Short atomic notes that link to each other, in preference to long ones. Planned pages (Rufus, 30 Oct):

  • Patterns of collective action (map of content): the two-level typology.
  • Names of collective action problems: a "Rosetta Stone" table of each game's standard name, its aliases by field, and the course name.
  • Prisoner's Dilemma, Stag hunt (update), Chicken (game), Volunteer's dilemma, Coordination game: one per game, each with its matrix, equilibria, aliases, typical failure and typical fix.
  • Greed and fear: the two questions that generate the map.
  • Chicken vs prisoner's dilemma, Prisoner's dilemma vs stag hunt, Stag hunt vs chicken: short comparison notes, each built on the matrices side by side.
  • Taking and building: the second axis (Ostrom's appropriation vs provision).
  • Aggregation technologies: best shot, weakest link, threshold (second level).
  • Principal-agent problem and Second-order free-rider problem: the second layer.

  Remaining red links are filed as Beads.

• KR5.4: The notes follow docs/wiki-conventions.md and docs/writing-style.md. An independent subagent checks the citations.

Sources: textbooks first, web last

Read the relevant chapters of a few good texts in full, from Rufus's library. Use web sources only to find page references or check a point, never as the main source.

Texts that matter for search, especially EPUBs, are converted to Markdown in the private library repo (~/src/life-itself/library/books/<slug>/, following its conventions) and referenced from there.

• Local now: Osborne and Rubinstein, A Course in Game Theory; Ostrom, Governing the Commons. Also whatever Rufus has since synced in Zotero.
• To obtain (course-wse.44):
  • Sandler, Collective Action (1992).
  • Dixit, Skeath and McAdams, Games of Strategy, the collective-action games chapter.
  • Mas-Colell, Whinston and Green, chapters 11, 13–14 and 23, or Laffont and Martimort.
  • Heckathorn 1996; Kollock 1998; Marwell and Oliver 1993; Alchian and Demsetz 1972.
• Classics (the relevant chapters, with page references): Olson 1965; Hardin 1982; Schelling 1960, and 1978 chapter 7; Ostrom 1990; Taylor 1987; Diekmann 1985; Hirshleifer 1983; Holmström 1982; Ullmann-Margalit 1977 (prisoner's-dilemma norms, coordination norms and norms of partiality, a close cousin of the top level).

Until the books arrive, pass 1 uses the local texts and standard results. Claims that rest on an unread source are marked [?] and rechecked when the book lands.

Practice follows docs/how-to-research.md. Working notes go in docs/plans/2026-10-29-typology-notes.md.

Scope

In: layer 1 (passes 1 and 2). Later: layer 2, the empirical literature (pass 3). Out:

• Hunting for new cases beyond course/research/.
• The evolutionary angle (Skyrms), beyond one line.
• A full survey of mechanism design.

Constraints: no edits to chapter scripts or slides; nothing committed or pushed without Rufus's go-ahead.

Done-check

Pass 1 is done when:

1. Every key result from KR1.1 to KR4.4 has an answer, or an explicit "not found" with where it was looked for.
2. Each top-level pattern has a payoff matrix and its equilibria.
3. The case-classification test has run, with orphans and double-fits listed.
4. Claims are cited with pages, or marked [?].
5. The results and decision sections below are filled in.

Pass 2 is done when the hub, the bibliography and the notes exist, pass the wiki conventions, and the citation check is clean.

Sign-off

• Rufus approves this brief (29 Oct, with the emphasis on a simple top level and the taking-or-building axis).
• Core reading obtained (course-wse.44).

Results

Pass 1 draft, 29 Oct. Full working, with payoff matrices, sources and the case test: 2026-10-29-typology-notes.md. Citation check: 2026-10-29-typology-citation-check.md.

Top level: two questions give three patterns. For any symmetric two-player game where cooperating together beats defecting together:

• Greed: if you pull your weight, do I gain by slacking?
• Fear: if I pull my weight alone, do I lose?

The map covers every symmetric 2×2 game where cooperating together beats defecting together. The rest, deadlock among them, are not dilemmas. Heckathorn 1996 (pp. 258–259) derives the same four games by swapping payoffs in the prisoner's dilemma. Pure coordination (which side of the road) is a one-line baseline.

Are chicken and the volunteer the same as the prisoner's dilemma? No. They differ by one swap: in chicken, defecting together is the worst outcome, not cooperating alone. That swap changes:

• The equilibria. The prisoner's dilemma has one, where nobody acts. Chicken and the volunteer have equilibria where someone acts, plus a mixed one where disaster has a real chance. In the volunteer's dilemma that chance grows with group size: the bystander effect.
• The failure. The wrong person pays, people wait, and there are commitment races.
• The fix. Allocate the job: name someone, rotate, pay. Assurance backfires: if I know you'll act, I wait. (This holds when one actor is enough. If the group is best off with everyone acting, chicken is fixed like a prisoner's dilemma: Dixit, Skeath and Reiley, 4th ed., p. 449.)

Two cautions from the 30 Oct recheck. In large groups, providing a public good almost always becomes a prisoner's dilemma (Dixit, Skeath and Reiley p. 430). Sandler's labels differ from ours: he calls the best-shot game a "coordination" game, though its payoffs are the volunteer's, and he gets chicken from goods that simply add up when doing nothing is a disaster (Collective Action, 1992, pp. 40–41; Global Collective Action, 2004, p. 68).

Rufus is partly right. The outcome looks the same (nothing gets done), real cases slide between patterns as the payoffs change, and repeated play blurs the line.

Second axis: taking or building, defined as restraint vs contribution. It rests on Ostrom's split between taking from a common resource and providing it (Governing the Commons, pp. 30–31, 47–50), simplified. In a simple matrix it doesn't change the equilibria, but the remedies differ:

• Taking needs limits, boundaries and monitoring.
• Building needs selective incentives, assurance and patrons.

Information commons are mostly building, with a taking side (enclosure, pollution).

Second layer.

• Principal-agent problems are adjacent: they involve two parties, not a group. They become a cause once a group delegates monitoring or enforcement ("who monitors the monitor").
• Moral hazard in teams and the second-order free-rider problem are battle lines one level down.
• Hold-up and lack of commitment are causes.
• Repetition, selective incentives, assurance contracts, Clarke–Groves and VCG, rotation and lotteries are remedies.

Case test.

• No pattern fits: Twelve Angry Men (an information cascade); the Aral Sea, the Ik, the Donner Party and Jaws (not collective action structures). Meerkats, and probably leaded petrol, aren't dilemmas at all ("no problem").
• Fit two patterns: about 13, including hoplites, the playground, Heartbleed, climate, the naval race, and the AI race (which fits all three). Almost all are many-player threshold cases, or cases whose payoffs shift with a parameter. That is a teaching point ("the pattern is in the payoffs, and payoffs move"), not a reason to add patterns.
• Chapter 1 grid: it marks taking × volunteer "(rare)", which is wrong. The Alanya fishing-site rotation (Ostrom p. 47) and Taylor's threshold fishery belong there.

Citation check. An independent subagent checked 70 claims:

• 53 confirmed.
• 1 wrong and 9 overstated. All are now fixed in the notes, and none changes a conclusion.
• 7 unverified.

All the matrices and equilibria were recomputed and are correct.

Rechecked 30 Oct against local copies of Olson, Sandler (1992, 2004), Dixit, Skeath and Reiley (4th ed.) and Schelling's Micromotives ch. 7: Olson's group types and "exploitation of the great by the small" confirmed with pages; Dixit's three forms confirmed; Sandler's mapping corrected as above; none of this changes the recommendation. Still not checked first-hand (marked [?] in the notes): Kollock 1998 beyond the abstract; Diekmann's original (the model is now page-cited from Dixit); Holmström and R. Hardin (seen only as quoted by Sandler); Alchian and Demsetz; Ullmann-Margalit; Marwell and Oliver. Schelling's page numbers come from a Google Books map and are unverified.

Decision for Rufus

Decided (Rufus, 30 Oct, after a Fable review): A.

• Course: three patterns, crossed with taking vs building:
  • Battle line (prisoner's dilemma);
  • Stag hunt (assurance game);
  • Chicken (volunteer's dilemma). "Chicken" replaces "volunteer" because people remember it. The bracket gives the many-player form, which is what group cases are.
• Fixes: enforce, reassure, decide who acts.
• Teaching: payoff matrices are shown and worked through; greed and fear are scaffolding, not the headline.
• Wiki: carries the full structure, with "no problem" as a real cell. A page mapping the double-fit cases, with a diagram, is proposed (Beads).

Options as put to Rufus:

How many patterns does chapter 1 teach?

Smaller question: keep "taking / building" as the grid's row labels, defined as restraint vs contribution? Recommend yes. Two row labels change.

Reply with A, B or C, plus yes or no on the row labels.

Input from the arc v0.7 review (30 Oct)

Arc v0.7 splits chapter 1 into types (the games, set by payoffs) and dimensions (five diagnostic questions: how many players, will we meet again, can defection be seen, can it be punished, is anyone big enough to act alone). Its critique is in the shared notes folder (2026-10-29-arc-v0.7-fable-critique.md). Its points, and how they fit this research:

• Keep types and dimensions apart. This matches the research. The greed/fear map classifies by payoffs only. Group size, repetition, visibility and the way contributions add up are second-level, which is the arc's dimensions.

• Two levers. The research supports naming both:

  • Changing the type means removing greed or fear: a fine turns a battle line into a stag hunt, and a refund guarantee removes fear in a stag hunt.
  • Changing the dimensions means repetition, visibility, punishment or a big player. These change the effective payoffs without changing the story.

  The hub page can label each fix as one or the other.

• Pure coordination and Schelling. Tie focal points to the stag hunt (picking the good equilibrium), and commitment to the battle line and chicken. Pure coordination stays a one-line baseline, not a fourth type.

• Decide once. Whatever Rufus picks, the arc, the ch1 script and this brief must all say the same.

• The arc review's lean is B: the volunteer's point is "is anyone big enough to act alone?", which is already a dimension. That is defensible. Dixit, Skeath and Reiley (p. 449) say allocation is chicken's distinctive fix only when one actor is enough, which is the same condition.

  Against it, still in favour of A:

  • The volunteer has different equilibria from both other games, so it is a distinct type, not a dimension.
  • "Anyone big enough" is only half of it. In an Olson privileged group with one big member, the problem disappears ("no problem"). It becomes a volunteer's dilemma only when several members are big enough and each would rather another paid (Olson p. 50, n. 70).
  • Standoffs and Strangelove need chicken anyway, in session 4.

  If Rufus prefers fewer concepts up front, B, with the volunteer as the example under "anyone big enough?", is the right form of B.