Use beliefs to explain why a better joint outcome may not be selected.
By the end: derive a coordination threshold and distinguish an announcement from an enforced agreement.
Two self-consistent outcomes
In Joint project, both joining yields 6 each. Choosing Safe guarantees 4 regardless of the other player's choice. A lone joiner gets 0. Both Join and both Safe are Nash equilibria. Joint Join is better for both, but that ranking alone does not select it.
A player uncertain about the other's willingness compares an expected payoff with the safe payoff. Let q be the probability that Column joins. Row's expected utility from Join is 6q; Safe gives 4. Join is strictly better when q exceeds two thirds, equal at two thirds, and worse below it.
Calculate the cost of being the only joiner
Coordination risk means that the appeal of an action depends on whether someone else takes a compatible action. In Joint project, choosing Safe guarantees Row four. Choosing Join gives Row six if Column joins and zero if Column chooses Safe. Relative to Safe, successful joining gains two and unsuccessful joining loses four. These differences explain why preferring the joint outcome is insufficient. Someone can value six more than four yet still reject a chance of ending at zero. The relevant comparison includes both the possible gain and the possible loss under that person's belief.
A probability is a number from zero to one describing the weight assigned to an uncertain event. A belief of fifty percent assigns weight one half to Column joining and one half to Column choosing Safe. Expected utility multiplies each possible own utility by its probability and adds the results. For Join at this belief, half of six contributes three and half of zero contributes zero. The total is three. Safe gives four in either case, so its weighted average remains four. The average describes a decision comparison, not an intermediate realized payoff.
The failed-join loss is twice the successful-join gain
Column choice
Join
Safe
Join advantage
Join
6
4
6 − 4 = 2
Safe
0
4
0 − 4 = −4
Half chance of each
3
4
½ × 2 − ½ × 4 = −1
Values compare Row's Join with Safe under the same opponent action. The larger failure loss explains why a fifty-percent belief is insufficient in this stipulated game.
Let q name Row's probability that Column joins. The chance of Column choosing Safe is then one minus q because those are the only two actions. Joining has expected utility six times q. Its advantage over Safe is six times q minus four. To find the break-even belief, ask what fraction of six equals four. Four divided by six is two thirds. Below that probability the advantage is negative; above it the advantage is positive. At exactly two thirds both choices maximize expected utility, so both are best responses rather than one being uniquely required.
An equivalent calculation starts with the gain of two from successful coordination and the loss of four from failed coordination. At a two-thirds chance of success, the weighted gain is four thirds. The one-third chance of failure gives a weighted loss of four thirds. They cancel. At a one-half belief, the weighted gain is one and the weighted loss is two, leaving a net disadvantage of one. This gain-and-loss method often transfers more easily than memorizing a threshold because it shows which consequence makes confidence necessary and how a changed consequence would affect the comparison.
The high threshold follows from these stipulated utilities, not from a general rule that cooperation requires two-thirds support. Row is choosing between a possible gain of two and a possible loss of four. A different safe alternative or a smaller loss would change that threshold. Nor do the numbers measure community welfare or cash earned by participants. They represent this player's preferences over the outcomes, with a numerical scale appropriate for expected-utility comparisons. Inferring the needed confidence in a real project requires defensible utility assumptions before applying the same arithmetic to observed behavior.
The exact crossing lies at two thirds
Both axes are quantitative. Join rises six utility units over the full probability range; Safe remains four. The crossing is between integer percentage control settings.
The two pure equilibria remain easy to verify without probabilities. Against certain Join, Join gives six rather than four; against certain Safe, Safe gives four rather than zero. Both players face these comparisons. Thus Join, Join and Safe, Safe are mutual best responses. The belief calculation fills the gap between certainty about either opponent action. It explains choosing Safe despite preferring joint Join. An efficient joint opportunity and an individually attractive uncertain choice are different objects with different comparisons.
A belief is not the other person's choice
The slider specifies Row's belief, not a measured rate or a command to Column. At q = 50%, Join averages 3, less than Safe's 4. At q = 100%, Join gives 6. The same preferences can support different best responses when beliefs differ.
A public standard or a trusted meeting point can help align expectations. A nonbinding announcement does not change the payoff table by itself. Whether people believe it depends on incentives, history, and information. An enforceable agreement instead changes available actions or their consequences and must be modeled separately.
Track whose belief is changing and what it means
Row's belief belongs to Row's decision problem. It need not equal Column's actual probability of joining, and Column may hold a different belief about Row. Suppose, as a paper example, Row assigns seventy-five percent to Column joining. Row's expected Join utility is four and a half, above Safe's four. Suppose Column assigns only fifty percent to Row joining. Column's expected Join utility is three, below four. Each player can best respond to their own stated belief, yet the resulting choices are Join, Safe, which is not a pure Nash equilibrium of the fixed game.
The mismatch arises because one person's forecast is not consistent with the other person's chosen action. Pure Nash reasoning asks whether each selected action best responds to the opponent's selected action. Subjective choice reasoning asks whether an action best responds to a player's belief. Both calculations can be performed correctly while answering different questions. To connect beliefs to an equilibrium prediction, you need additional consistency assumptions about expectations and strategies. The coordination slider supplies only Row's belief. It does not specify a second belief, calculate a belief-consistency condition, or observe either player's behavior.
Two subjective best responses need not fit together
Row believes 75% Join
Utility
Join
0.75 × 6 = 4.5
Safe
4
Choice
Join
Column believes 50% Join
Utility
Join
0.5 × 6 = 3
Safe
4
Choice
Safe
Paper beliefs only. Row expects joining more strongly than Column does. Each optimizes against a different forecast; the resulting pure profile is not a Nash equilibrium.
Use the Uncertain preset to set the displayed belief to fifty percent. Confirm that Join gives three while Safe stays at four. Move the slider to sixty-six percent. Join becomes three point nine six because six times sixty-six hundredths is three point nine six. Move to sixty-seven percent and Join becomes four point zero two. The preferred action changes between those adjacent settings. The exact equality is at sixty-six and two-thirds percent, between integer slider values.
The plot keeps both utility lines visible because the decision depends on their difference. At zero percent, Join is zero and Safe is four. At one hundred percent, Join is six and Safe is four. Safe's horizontal line records that its utility is independent of Column's action in this particular game. It does not mean Safe is free of every real-world risk. Such risks were excluded when the utilities were specified. Changing the belief control changes neither the safe payoff nor the utility received by a lone joiner or a successful pair.
The labels Certain Safe and Certain Join are beliefs about Column, not buttons that choose Column's action. Moving between them does not make a project succeed or fail. Likewise, the middle belief is not a claim that half the neighborhood has signed up. A measured participation fraction would require observations, a defined population, and a reason that the fraction predicts this opponent's choice. An observer's statistical estimate and a decision-maker's subjective probability may coincide, but the interface does not establish that connection. Read the numbers as conditional reasoning under a supplied belief.
Adjacent slider settings straddle the exact tie
Belief Column joins
Join utility
Safe utility
Best response
66%
6 × 0.66 = 3.96
4
Safe
Exactly 66⅔%
6 × ⅔ = 4
4
Both
67%
6 × 0.67 = 4.02
4
Join
The exact threshold is a mathematical comparison, not a selectable preset. Integer percentage steps show strict preferences on either side without changing the utility table.
Separate uncertainty about another person's action from uncertainty about the game's consequences. Here Row knows that joint Join yields six and lone Join yields zero, but does not know Column's choice. If the project might fail even when both join, the successful-join payoff must itself be evaluated under that additional uncertainty. If Column has different preferences, Row may need a different explanation for the belief. Neither extension is controlled in this lab. Keeping the source of uncertainty explicit prevents a participation forecast from being mistaken for an engineering reliability estimate or a verified payoff measurement.
Compare incentives and outcomes
Compare the project with the safe option
Joint project. Each cell lists Row, then Column utility.
Row / Column
Join
Safe
Join
Row: 6 ✓Column: 6 ✓Both best responses
Row: 0Column: 4
Safe
Row: 4Column: 0
Row: 4 ✓Column: 4 ✓Both best responses
✓ means that player cannot improve by switching alone while the other action stays fixed. Ties receive marks too. A border identifies the selected cell.
Solid: JoinDashed: Safe
Join: 3. Safe: 4. Best response: Safe.
Find the break-even belief
q is the probability that Column joins. The 6 and 4 are Row’s utility values.
The slider changes only Row’s belief about Column. The game remains fixed. This is an expected-utility comparison, not an estimated participation probability. Join and Safe tie at exactly two thirds, between the slider’s integer percentage values.
Belief Column joins
Expected Join
Safe
1/2
3
4
2/3
4
4
1
6
4
These are expected utility units for Row, not realized attendance.
Compare changes to rewards, losses, and the safe option
The following are paper variations, not editable scenarios in the lab. Start from Row's Join payoffs six against Join and zero against Safe, with Safe fixed at four. Raising the joint-join payoff to eight gives expected utility eight times q. Equality with four now requires q equal to one half. A larger successful reward makes participation attractive at lower confidence. Column's response to the reward remains unspecified.
Instead retain the successful payoff of six and give a lone joiner two rather than zero. Expected Join utility becomes six times q plus two times one minus q. The first part rewards successful joining; the second protects unsuccessful joining. Combining the terms gives two plus four times q. Equality with Safe's four again occurs at q equal to one half. This protection changes a different outcome than the larger reward, even though the two variations share a threshold. At certainty of success their utilities differ, so matching thresholds do not make the policies identical.
Equal thresholds can conceal different consequences
Row utility rule
At q = 0
At q = ½
At q = 1
Threshold
Baseline Join: 6q; Safe 4
0
3
6
⅔
Reward Join: 8q; Safe 4
0
4
8
½
Protected Join: 2 + 4q; Safe 4
2
4
6
½
Worse Safe: 6q; Safe 2
0
3
6
⅓
All alternatives to the baseline are paper variations, not controls. Reward and protection each lower the threshold to one half, but give different utilities at other beliefs.
Reducing Safe from four to two while retaining Join payoffs six and zero gives a threshold of one third. This makes Join preferable at more beliefs by worsening the alternative, not by improving the project. At a belief of one half, the original best attainable utility was four through Safe. Under the reduced-safe variation it is three through Join. Increased participation can therefore coexist with a lower maximum expected utility for the chooser. Calling every threshold reduction beneficial would confuse a changed action with an improvement in the person's available outcomes or an established social objective.
For a general arithmetic method, call the successful Join utility a, the failed Join utility b, and the constant Safe utility s. Assume a exceeds s and s exceeds b. Starting at b, each unit of probability moved toward success adds a minus b to expected utility. Reaching s therefore requires probability equal to s minus b divided by a minus b. In the baseline, the required improvement is four and the full possible improvement is six, giving two thirds. This is a ratio of utility differences within one person's scale, not a comparison between people.
Check the ordering assumptions before using that ratio as a meaningful interior threshold. If Safe guaranteed seven while Join still gave six or zero, Join would be strictly worse against both opponent choices. The formal ratio seven divided by six exceeds one, indicating no possible belief makes Join preferable. If Safe instead gave zero, Join would tie at a zero belief and improve at every positive belief. If successful and failed Join had equal utilities, there would be no probability slope to divide by. These failures identify different games rather than numerical nuisances to hide by rounding.
Check the payoff ordering before interpreting a threshold
Safe guarantee
Against certain Safe
Against certain Join
Belief implication
7
Join 0 < 7
Join 6 < 7
Join never optimal
4
Join 0 < 4
Join 6 > 4
Interior threshold ⅔
0
Join 0 = 0
Join 6 > 0
Tie at 0%; Join above 0%
Paper comparisons keep Join at six against Join and zero against Safe. An apparent probability above one means no feasible belief supports joining, not extreme confidence.
A real guarantee needs its own accounting. Someone must supply the lone-joiner protection, and the promise must remain enforceable when coordination fails. If compensation depends on the very project that failed, the stipulated utility of two may be unjustified. If a third party pays, that party is absent from the two-player welfare comparison. Keep the fixed-game slider for belief changes and use the paper comparisons for changed consequences, so an explanation never attributes a payoff intervention to mere optimism.
Coordination and conflict can coexist
Two people can agree that meeting is better than missing each other yet prefer different venues. Such a game has a distributional conflict between coordinated outcomes. A model with identical preferences over the coordinated cells omits that conflict.
Ask who bears the cost of failed coordination, whether a safe action is genuinely safe, and what evidence supports the belief. A mathematically efficient project may need an institution that makes participation credible. Do not infer that coordination failure proves laziness or ignorance.
Ask whether a message aligns expectations or changes incentives
A recommendation can matter in Joint project because Join is already a best response to Join. Suppose both players receive a public suggestion to join and each becomes certain that the other will follow it. Each then compares six with four and wants to follow. The message has aligned expectations on an existing equilibrium without changing the matrix. If instead each expects the other to ignore the recommendation and choose Safe, Safe remains optimal. The content of a recommendation alone supplies no arithmetic reason to replace one expectation with the other or a guarantee that beliefs actually change.
Compare the unchanged maintenance game from earlier. Even if a message makes Row certain that Column will Maintain, Row receives five by skipping instead of three by maintaining. Certainty about the other person's cooperation does not support mutual maintenance under those one-shot preferences. This contrast explains when communication might help with equilibrium selection and when it leaves a direct incentive to deviate. People might still respond to messages. A message might reveal preferences or introduce other consequences, but those mechanisms would need their own model and evidence.
Certainty about cooperation helps only if cooperation best responds
Joint project
Row utility
Join against Join
6
Safe against Join
4
Best response
Join
Shared maintenance
Row utility
Maintain against Maintain
3
Skip against Maintain
5
Best response
Skip
Compare the actual presets under certainty about the cooperative opponent action. A recommendation changes no payoff, so only Joint project already supports compliance against compliance.
Coordination can also include disagreement over the preferred coordinated outcome. Consider a paper venue game, not a preset. Meeting at East gives Row six and Column four. Meeting at West gives Row four and Column six. Missing each other gives both zero. Both coordinated profiles are Nash equilibria because switching alone loses the meeting payoff. Neither coordinated profile Pareto-dominates the other. A recommendation of East therefore asks Column to accept its less preferred equilibrium, while a recommendation of West asks the same of Row. Shared interest in meeting does not remove the distributional conflict.
At a fifty-percent belief that Column chooses East, Row's expected utilities are three for East and two for West, so Row prefers East. At a fifty-percent belief that Row chooses East, Column's expected utilities are two for East and three for West, so Column prefers West. The resulting East, West profile misses. A communication procedure must address both expectation alignment and conflicting preferences. Simply repeating that meeting is good does not resolve which meeting to choose.
An enforceable agreement differs from a recommendation because it changes the consequences or permitted actions. A penalty on choosing Safe could remove Safe, Safe as an equilibrium, but only after its size, enforcement, and recipient are specified. A promise of future punishment would require a longer game and credible future behavior. Neither mechanism is present in the coordination slider. The later commitment lesson specifies enforcement. Here distinguish changing the game from selecting an existing equilibrium.
Both want to meet, but they prefer different meetings
Row / Column
East
West
East
6, 4
0, 0
West
0, 0
4, 6
At half-half belief
East utility
West utility
Row
3
2
Column
2
3
Separate paper game, not a preset. Both diagonal cells are equilibria. At half-half beliefs, Row prefers East and Column prefers West, producing an uncoordinated action pair.
For a transferable diagnosis, first compare each person's alternatives under the same opponent choice. Then identify the belief needed for the proposed action and the evidence that might support that belief. If everyone prefers the same coordinated outcome but fears being alone, coordination risk is a plausible mechanism. If players prefer different coordinated outcomes, distribution also matters. If someone still prefers to deviate when certain others comply, expectation alignment alone is insufficient. These conditional diagnoses narrow what an intervention must accomplish without claiming that one small matrix forecasts participation, establishes fairness, or measures the effects of a real institution.
Exercises
Change the safe alternative
Suppose joint joining still gives 6 and joining alone gives 0, but Safe now guarantees 2. What is the new threshold?
Hint
Set the expected joining payoff equal to 2.
Worked answer
Safe payoff
Break-even belief
Reason
4
2/3
6 × 2/3 = 4
2
1/3
6 × 1/3 = 2
A payoff change moves the threshold, not the evidence about the opponent.
The break-even condition is 6q = 2, so q = one third. Lowering the safe payoff makes Join preferable at more beliefs. It does not establish that Column is more likely to join, and a worse alternative is not automatically a socially desirable policy.
An outcome selector
A coordinator recommends Join. Does this eliminate Safe, Safe as a Nash equilibrium in the unchanged game?
Hint
Distinguish a message from a changed payoff or action set.
Worked answer
Intervention
Matrix changes?
What needs evidence?
Recommendation only
No
Whether it changes beliefs and choices
Enforced penalty
Yes
Enforcement, amount, and new incentives
Equilibrium selection and changing the game are distinct.
No. If the payoffs and available actions remain unchanged, Safe is still a best response to Safe. The recommendation may influence beliefs or behavior, but it does not remove that equilibrium from the original matrix. Enforced commitments or penalties would define a different game.
Sources and further reading
Selected reference passages support the definitions and algorithms below. The games, explanations, and implementations are original. Terminology and assumptions are stated within each lesson; source review is not an independent audit of this course. Optional links are not runtime dependencies.