Interdependence Theory: An S-Tier Behavioral Designer’s Guide

Two players sit down at a table, each with two buttons. The buttons don’t cost anything to press. The buttons don’t do anything by themselves. They only resolve into outcomes when the other person presses theirs. That table — the one where my outcome is a function of what you do, and yours is a function of what I do — is the entire subject of interdependence theory. Every multiplayer feature, every team product, every pricing tier with a referral kicker, every cooperative quest, every shared inbox, every couples’ therapy session is just a different table laid on top of the same matrix.

If you have ever shipped a feature where the user’s experience depends on what someone else does — a friend, a teammate, a spouse, a stranger in a Discord — you have shipped a structure that interdependence theory is older than, sharper than, and more honest about than the framing in your spec. The spec said “encourage cooperation.” Kelley and Thibaut, writing in 1959 and finishing the formal version in 1978, would have asked you the question that actually matters: which outcome matrix did you ship?

This is a guide to the answer.

Speed Run Notes

• Interdependence theory replaces “motivation” with a 2×2 outcome matrix: my payoff is a function of my choice and yours. Every cooperation problem is a matrix-shape problem.
• The given matrix is what the situation literally pays out. The effective matrix is what people actually decide on after re-weighting outcomes for the partner’s welfare. Most design lives in the gap between the two.
• Kelley & Thibaut isolated six dimensions that classify every interdependent situation: degree, mutuality, covariation, basis (exchange vs coordination), temporal structure, and information availability.
• The 1980 Investment Model (Caryl Rusbult) made the theory testable in real relationships: Commitment = Satisfaction + Investment − Quality of Alternatives. Decades of replication followed.
• Rusbult’s accommodation work showed that “loving partners” differ from “short-term partners” less by feeling and more by which transformation rule they apply to provocations.
• For Octalysis designers, the matrix is an X-ray of CD5 mechanics: it tells you whether a feature is exchange (zero-sum), coordination (we win together), or mixed-motive (everyone’s personal incentive points away from the joint optimum).

About the Author

Yu-kai Chou is an S-Tier Behavioral Designer and the creator of the Octalysis Framework, the gamification design system now applied to products and experiences reaching over 1.5 billion users. His book Actionable Gamification is one of the most-cited works in the field, and he has been ranked the #1 Gamification Guru in the World.

He has advised MrBeast, LEGO, Microsoft, Porsche, Tesla, Stanford, Harvard, and governments including Ukraine on turning behavioral psychology into product mechanics that actually change user behavior.

Verify: Wikipedia · Google Scholar · Wikidata · LinkedIn

I’ve spent the better part of two decades building Octalysis around the question of how Core Drive 5 (Social Influence & Relatedness) actually plays out inside a product, and the more multiplayer features I’ve consulted on — for studios, for fintech apps, for K-12 learning platforms, for HR teams designing hybrid offices — the more I’ve come back to interdependence theory as the substrate. Octalysis tells you which feeling to engineer. The Kelley – Thibaut matrix tells you which structural variant of a multiplayer interaction you’re actually shipping. Without the matrix you can ask the audience to be more cooperative, but you can’t tell them which table you’ve sat them down at. This guide is the version I use in workshops.

What is Interdependence Theory

Interdependence theory is a formal framework for analyzing any situation in which two or more people’s outcomes depend on each other’s choices. Harold Kelley and John Thibaut developed it in two stages — the 1959 book The Social Psychology of Groups introduced the outcome-matrix vocabulary, and the 1978 follow-up Interpersonal Relations: A Theory of Interdependence formalized the six dimensions and the “effective matrix” concept that lets behavior diverge from naked self-interest.

The starting move is deceptively boring: take any social situation and represent it as a payoff matrix. Two players, each with two options, four cells in the matrix, each cell containing the payoffs both players receive when those two options are chosen. The Prisoner’s Dilemma is one matrix. Buying a coffee is another. Sharing a meal at a restaurant where the bill will be split evenly is a third. A teacher deciding whether to write a generous letter of recommendation for a student is a fourth. Once you have the matrix, you can read off everything that matters: how interdependent the two parties really are, who controls whose payoffs, whether their interests align or oppose, and what the game would tell a purely self-interested actor to do.

That “tell a purely self-interested actor to do” part is where the theory becomes interesting. Kelley and Thibaut never claimed people choose like rational economic actors do — they claimed people start from the given matrix (what the situation pays) and apply a transformation to it (what their values, attachment to the partner, identity, norms, and relationship history say to weight). The output of that transformation is the effective matrix. The effective matrix is what they actually decide on. The whole project is the science of those transformations.

This sets interdependence theory apart from any motivation theory whose grammar is “person wants X.” Interdependence theory’s grammar is “person and partner are sitting in a particular structural arrangement, and that arrangement — before any motive is invoked — constrains and enables the moves they can make.” You don’t fix a low-cooperation game by exhorting people. You fix it by changing the matrix.

The Core Findings

Across roughly six decades of work in the Kelley – Thibaut tradition — including Caryl Rusbult’s extensions, Paul Van Lange’s social value orientation research, and Margaret Clark and Judson Mills’s exchange — communal distinction — a small number of findings have hardened into the core of the theory.

The 2×2 outcome matrix is generative

Almost every meaningful social situation can be decomposed into a small set of canonical matrix shapes. The Prisoner’s Dilemma, Chicken, Battle of the Sexes, Stag Hunt, Threat games, and Trust games all show up in everyday life. They are not academic abstractions. The matrix shape predicts which moves will be tempting, which will feel costly, and which will require something like trust or commitment to overcome.

Six dimensions classify the matrix

Kelley and his collaborators identified six structural properties that, together, locate any interaction in a structural taxonomy: degree of dependence (how much do my outcomes vary with the partner’s choice), mutuality (do we depend equally on each other), correspondence vs conflict of interest (do our payoffs covary positively or negatively), basis of dependence (exchange vs joint coordination), temporal structure (one-shot, sequential, or extended), and information availability (do I know your payoffs and your moves). Get those six right and you have characterized the situation.

The effective matrix differs from the given matrix

People do not maximize the given matrix. They first transform it — weighting their partner’s outcomes alongside their own, reframing competitive stakes as cooperative ones, applying long-run rather than short-run accounting. The transformation rule a person reaches for is shaped by their disposition (Van Lange’s Social Value Orientation work shows people stably differ between prosocial, individualistic, and competitive orientations), by the relationship’s history (commitment shifts the rule), and by the situation itself (anonymity tends to suppress prosocial transformations).

Commitment is the integrator variable

Rusbult’s 1980 Investment Model is the most empirically fertile descendant of the parent theory. It says commitment to a relationship is a function of three inputs: satisfaction (does the relationship deliver good outcomes), quality of alternatives (could you do better elsewhere), and investment (how much have you sunk in that you cannot extract). Commitment, in turn, predicts persistence behaviors — accommodation in the face of provocation, willingness to sacrifice, derogation of attractive alternatives, forgiveness. The model has been tested on romantic couples, employees and organizations, athletes and teams, voters and parties, even users and platforms.

Accommodation is the relationship-grade behavior

When a partner does something destructive — complains, criticizes, withdraws — people choose between four classes of response that Rusbult’s framework labels Exit, Voice, Loyalty, and Neglect. Highly committed partners disproportionately choose constructive responses (Voice and Loyalty) over destructive ones (Exit and Neglect), even when the destructive responses are more “rational” in a one-shot accounting. That asymmetry is accommodation, and it is a behavioral signature of the commitment transformation.

Communal vs exchange relationships are governed by different rules

Margaret Clark and Judson Mills’s work distinguishes communal relationships, in which benefits are given in response to need or to please the partner, from exchange relationships, in which benefits are tracked and reciprocated like a ledger. Importing the wrong rule into the wrong relationship type is a reliable source of conflict: keeping a tally with a close friend reads as cold; refusing to track payments with a coworker reads as exploitative.

What Kelley & Thibaut Got Right

The theory has aged better than most of its mid-century social-psychology cousins. Three things they got right deserve specific credit.

They made the situation a first-class citizen. Personality psychology in the 1950s was still dominated by trait-and-need explanations — people behaved a certain way because of an enduring disposition. Kelley and Thibaut insisted on the matrix first, the person second. The same person sitting in a Prisoner’s Dilemma matrix and a Stag Hunt matrix will make different choices not because their personality changed but because the structural shape of the payoffs changed. This insight prefigured Mischel’s person × situation revolution by a decade and is the reason the theory is still load-bearing in modern relationship science.

They formalized cooperation without assuming altruism. By distinguishing the given matrix from the effective matrix, the theory could explain why a self-interested person sometimes acts in a partner’s interest without postulating either irrational sacrifice or hidden-self-interest cynicism. Both extremes are tempting and both miss the mechanism. The transformation rule sits in the middle — a real psychological process that converts “what I would do as a stranger” into “what I will do here, given who I am to this person.”

They built a framework that downstream theories could plug into. Investment Model, Social Value Orientation, accommodation, communal/exchange, the Atlas of Interpersonal Situations — none of these are restatements of the parent theory; they are new pieces snapped onto its frame. That kind of generativity is the signature of a theory that captured something real, not just a useful vocabulary.

They were honest about the prediction problem. Both 1959 and 1978 are unusually candid about what the theory does and does not predict. The 1978 book repeatedly says the matrix taxonomy is a vocabulary for analyzing the structural arrangement of interdependent situations — it is not, by itself, a generator of behavioral predictions. Predictions arise only when the theory is paired with a model of the transformation rule the actors will apply. That epistemic modesty is rare in mid-century social psychology and is why the theory has aged better than its more confident contemporaries.

Where Interdependence Theory Falls Apart

For all its strengths, the theory has structural weaknesses that any honest designer should hold alongside the framework. These are not boilerplate caveats — they are the parts that, once you see them, change how you use the matrix.

The matrix is a 2×2 fiction in a 4D world

Real interdependence rarely cleanly decomposes into two players, two choices, and four cells. A team of seven on a shared deadline is not a 2×2 anything. A platform with millions of users tracking each other’s likes is governed by network-level dynamics that no dyadic matrix captures. When Kelley and his collaborators tried to extend the theory to triads and larger groups, the analytical tractability that made the dyadic version powerful collapsed under the combinatorial explosion. The theory is most rigorous exactly where most modern social design isn’t — one-on-one. Designers who reach for the matrix for a 50-person guild structure are using a borrowed metaphor, not the theory’s actual machinery.

The transformation rule is a black box

The theory says people transform the given matrix into an effective matrix using a rule, and the rule depends on disposition, history, and context. That is true. It is also not a mechanism. Forty years on, we have no clean way to predict in advance which transformation a given person will apply to a given matrix. Van Lange’s SVO measure helps, attachment style helps, commitment level helps — but the residual variance is enormous, and the “effective matrix” remains, in practice, a thing we can only reconstruct after the fact from observed behavior. That is uncomfortably close to a tautology: the effective matrix is whatever explains what the person did, because what the person did is what the effective matrix predicted. Useful framing, weak prediction.

The Investment Model has a measurement — not just measurement — problem

Rusbult’s Investment Model has been replicated extensively, but the replications mostly use the Investment Model Scale, which asks participants to rate their own satisfaction, alternatives, and investment on Likert items. That is a self-report measure of three constructs that themselves are partially defined by their predictive relationship with commitment. The 2010 meta-analysis by Le and colleagues confirmed the model’s structure across hundreds of samples but acknowledged that effect sizes shrink in studies that use behavioral rather than self-report outcomes. The model predicts what people say about their commitment more reliably than what they do about it — and the gap between those two has eaten more relationship products than any other modeling assumption.

The deeper problem is partial circularity. “Investment” is operationalized as “things you would lose if you left” — but whether something is felt as a loss is itself partly a function of how committed you already are. A user who is on the fence about quitting a fitness app sees their accumulated workout streak as a sunk cost they’d hate to lose; a user who has already mentally exited sees the same streak as data they no longer care about. The Investment Model does not predict which users will read which way. It works best as a snapshot diagnostic on a population, not as a leading indicator on an individual.

Cultural specificity of the matrix logic

The matrix vocabulary, the “rational starting point” of the given matrix, and the prosocial-as-deviation framing of the transformation rule are recognizably products of a particular intellectual tradition — mid-century American behaviorism filtered through game theory. Cross-cultural replications have shown that the Investment Model’s three components hold up reasonably well across collectivist and individualist samples, but the weights shift dramatically. In tighter cultures (Gelfand’s tightness-looseness work) investment dominates; in looser ones, alternatives. In societies with strong third-party commitments built into the relationship (extended family, kinship obligations, religious frameworks), the matrix is rarely dyadic in the first place — the partner’s family is sitting at the table even when only two people are in the room. A theory built on isolated dyadic interactions struggles to capture that.

The Brain on Interdependence

Twenty-five years of social neuroscience has fleshed out what was originally a behaviorist’s vocabulary with neural correlates that, in places, beautifully match the theory and in others awkwardly contradict it.

The medial prefrontal cortex (mPFC), and especially its dorsomedial subdivision, lights up reliably when people consider another person’s outcomes alongside their own. The temporoparietal junction (TPJ) handles perspective-taking — the “what is the partner seeing” computation that any matrix transformation requires as a precondition. The ventral striatum, originally identified as a reward-processing region, also tracks rewards received by liked others, providing a candidate neural substrate for the prosocial transformation rule. When the partner is disliked or out-group, the ventral striatum response shifts toward Schadenfreude territory, with the dorsal anterior cingulate cortex (dACC) flagging conflict.

The clearest brain-level evidence for accommodation comes from work on conflict-reappraisal in close relationships. When highly committed partners are exposed to a partner’s provocation, fMRI studies show down-regulation of insula and amygdala activity coincident with up-regulation of ventrolateral prefrontal cortex — the canonical signature of emotion regulation. Less committed partners show the opposite pattern: amygdala up, prefrontal down, exit-or-neglect responses follow. The transformation isn’t just behavioral; it’s a regulated emotional response that recruits the same circuitry as cognitive reappraisal.

Where the picture gets messier is in trying to localize the “effective matrix” itself. Neuroimaging cannot show us a brain region that is the transformed matrix. What it shows is a network — mPFC, TPJ, ventral striatum, posterior cingulate — that comes online whenever the social-decision context is rich enough to require considering a partner’s perspective. That network’s activity correlates with prosocial choices, but it doesn’t prove that the brain explicitly represents and computes a matrix. The matrix may be the right description for the behavior; the brain may be doing something more like a soft, continuous gradient computation that approximates matrix logic without instantiating it.

Two practical implications fall out of the neural picture. First, anything that taxes the prefrontal regulation system — sleep deprivation, alcohol, cognitive load, time pressure — reliably degrades the prosocial transformation. The matrix the tired version of you plays is closer to the given matrix than the matrix the rested version plays. Designers building features that mediate emotionally charged interactions (apologies, complaints, negotiations) should consider that load deliberately. Second, repeated successful prosocial choices in a relationship appear to strengthen the connectivity between mPFC and the regulation network — the transformation rule is not just a setting but a trained capacity. Long-term partnerships, in this picture, literally rewire the participants’ ability to play matrix games together.

Interdependence Theory vs Other Theories

vs Social Exchange Theory. Both descend from a common Thibaut & Kelley taproot, but social exchange theory (Homans, Blau) keeps the “rational ledger” framing — people maintain relationships when rewards exceed costs, full stop. Interdependence theory accepts that as a starting point and then adds the transformation rule, which is exactly what social exchange theory cannot account for. If exchange theory predicts a person will leave a relationship the moment alternatives improve, interdependence theory predicts a committed person won’t, because their effective matrix has down-weighted the alternatives. The Investment Model’s “commitment buffers against alternatives” finding is a direct empirical refutation of pure exchange theory.

vs Equity Theory. Adams’s equity theory predicts dissatisfaction whenever the input/output ratio is unequal between partners. Interdependence theory says equity is one possible transformation rule, and which rule applies depends on the relationship type. Communal partners don’t track equity moment-to-moment. Exchange partners do. Both can be stable. Equity theory, by contrast, predicts that all unequal relationships will fail.

vs Game Theory. Game theory and interdependence theory share the matrix vocabulary but answer different questions. Game theory asks what a rational actor with stated preferences should do given a matrix — the prescriptive question. Interdependence theory asks what a real person, with messy and other-oriented preferences, will do — the descriptive psychological question. Game theorists treat the matrix as exogenously given. Interdependence theorists treat the given matrix as one input to a transformation that produces the matrix the person actually plays.

vs Self-Determination Theory. SDT (Deci & Ryan) lives at the level of the individual’s three innate psychological needs: autonomy, competence, relatedness. Interdependence theory lives at the level of the dyad’s structural arrangement. They’re complementary — SDT can tell you a person needs relatedness; interdependence theory can tell you which matrix shape will deliver it and which will starve it. The cleanest design pattern is to use SDT as the diagnostic and interdependence theory as the structural prescription.

vs Attachment Theory. Attachment theory describes how early caregiver experiences leave templates that shape adult relationships — secure, anxious, avoidant. Interdependence theory describes the structural arrangements that shape any given interaction. Attachment is the disposition you bring; the matrix is the table you sit down at. The two combine at the transformation step: an avoidantly attached person facing a Stag Hunt is more likely to defect to the safer hare than a securely attached person, holding the matrix constant.

vs Octalysis. Octalysis tells you which of eight Core Drives a feature engages and at what intensity. Interdependence theory tells you the structural shape of the social situation that hosts that feature. Most CD5 (Social Influence & Relatedness) techniques can be implemented across multiple matrix shapes — the technique is the same, the structure underneath it is different, and the resulting behavior is wildly different. A leaderboard implemented as an exchange matrix breeds toxicity; the same leaderboard as a coordination matrix breeds collaboration. Octalysis tells you to pull the lever; interdependence theory tells you which lever you actually have.

Interdependence Theory in the Real World

Romantic relationships and the Investment Model

The biggest body of empirical interdependence-theory work is in romantic relationships. The Investment Model predicts — with effect sizes that have held up across cultures from US college samples to Indian arranged-marriage samples — that satisfaction, alternatives, and investment together account for 50–70% of variance in commitment, and commitment in turn predicts breakup probability over 6 to 36-month follow-ups better than satisfaction alone. The practical translation: a partner who is unhappy but heavily invested and sees no good alternatives is more likely to stay than a partner who is moderately happy but has attractive options. Couples therapy that ignores any of those three levers is doing partial work.

Workplace teams and stack-ranking

Stack-ranking systems — where employees are forced into bell-curve distributions and the bottom is regularly culled — are an exchange matrix wearing a coordination matrix’s costume. The org wants the team to function as a Stag Hunt (we win together) but the formal incentive structure is a Prisoner’s Dilemma (defect on your colleague to avoid being culled). The structural mismatch is why stack-ranking reliably destroys the trust it claims to enforce. Microsoft killed its stack-ranking system in 2013 after years of behavioral evidence that engineers were withholding code from teammates and gaming review cycles — behavior that the matrix predicts and that exhortations to “collaborate more” could never override.

Multiplayer game design

Cooperative games (Pandemic, It Takes Two, Overcooked) are matrices designed so that defection is mathematically impossible — both players win or both lose. Competitive games (chess, Counter-Strike) are matrices where defection isn’t even available because the moves all are “defection” against the opponent. The interesting design space is mixed-motive games — Diplomacy, EVE Online, every social-deduction game — where the matrix is structured so that some players sometimes have an incentive to defect on collaborators they will work with again. The strategic depth players love comes directly from the unresolved tension between given and effective matrices, which means designers raise stakes by widening that gap, not by closing it.

Loyalty programs and switching costs

Every loyalty program that successfully retains users is, structurally, an attempt to manipulate the three Investment Model variables. Status tiers raise satisfaction (recognition feeds CD5). Earned points raise investment (sunk-cost-feeling under CD4 Ownership). Exclusive partnerships and benefits suppress alternatives (everyone else looks worse by comparison). Programs that lean only on points are doing one-third of the job. Airlines that combine all three — Delta’s SkyMiles with status tiers and Delta-only premium routes — outperform pure-points programs in retention by 20-40 percentage points across published industry analyses.

Online dating and the alternatives explosion

The Investment Model predicts an unhappy implication for any platform that radically increases the visibility of alternatives: holding satisfaction and investment constant, commitment will fall. That is exactly what dating-app studies have found. Tinder’s near-infinite swipe queue makes any current partner’s flaws more salient because the “next option” is one swipe away. Apps that have tried to push the other direction — Hinge with its “designed to be deleted” positioning, Coffee Meets Bagel with its single curated daily match — are explicitly suppressing alternatives, an interdependence-theory-savvy design choice that reduces the very platform feature competitors lean on.

Education and the group-project paradox

Every teacher who has tried to grade group projects has run into the structural problem interdependence theory diagnoses: the project itself is a Stag Hunt (we all coordinate or we all fail), but the grading rubric — one shared grade for the team — turns it into a Prisoner’s Dilemma where the rational individual move is to free-ride on the most conscientious teammate. Decades of pedagogy research, summarized by Johnson and Johnson’s cooperative-learning framework, converge on the same fix: pair the joint outcome with individual accountability so that defection is detectable and punishable at the individual level. The interdependence-theory translation is that you have to ship both the inner cooperative matrix and the outer accountability wrapper at the same time. Educators who only ship the first watch the diligent students burn out; those who only ship the second get rugged individualism, not collaboration.

Open-source software contribution

Open-source projects are an instructive natural experiment in how interdependence structure shapes long-term collaboration. The Linux kernel, with its strong maintainer hierarchy, public review history, and reputation-driven contribution graph, is structurally a coordination matrix wrapped in a heavy reputational system. Anyone can submit a patch but their identity follows them across years of commits. The reputation wrapper is what stops free-riding from collapsing the whole thing — even though no contributor is paid by the kernel project itself. Compare this to anonymous code-paste sites where the matrix is one-shot and the reputation wrapper is absent: the equilibrium drops to the lowest-quality output the matrix can sustain. Same Octalysis Core Drives engaged (CD3, CD5, CD2), wildly different long-run outcomes, because the matrix temporal-structure dimension is different.

The Elephant in the Room

Interdependence theory’s biggest blind spot, almost never named in the textbooks, is that the matrix vocabulary makes cooperative outcomes feel like the natural success state and competitive ones feel like a failure mode that better matrix design should eliminate. That bias has been baked into so many cooperative-design rhetorics that it now reads as common sense.

It’s also wrong. Plenty of healthy human interaction is structurally competitive — sports, courtship, market exchange, debate, the entire creative-discipline tradition of competitive critique — and the goal isn’t to redesign the matrix into cooperation. The goal is to set the competitive matrix inside a higher-level cooperative wrapper (we both want a fair game; we both want the institution that hosts the competition to thrive) so that defection at the meta-level — cheating, sabotage, coalition-busting — is what cooperation looks like at the higher level. Sports leagues spend roughly as much energy on integrity policing as they do on staging the games themselves, because the wrapper is what makes the game worth playing.

Designers who only know the theory’s “turn defection matrices into coordination matrices” move will end up either flattening interactions that needed competitive juice (most cooperative-only games are forgotten the year they ship) or smuggling competition back in through the back door without the integrity infrastructure to host it (every “cooperative” ranked queue that breeds toxicity). The honest version of the theory recognizes that the design question is rarely “cooperation or competition” — it’s “at which level of the nested matrix do I want each.”

How to Apply Interdependence Theory with the Octalysis Framework

The translation of interdependence theory into Octalysis is the design move that most consistently rescues failing CD5 features, because it turns “our community feels off” into “our matrix is a Prisoner’s Dilemma when it should be a Stag Hunt.”

Diagnose the matrix shape before picking a Game Technique

Before you reach for a CD5 technique — Group Quest, Mentorship, Friending, Social Treasures — sketch the actual matrix the feature creates. Two players, two main choices each, four cells. What does each player get when they cooperate? What does each get when they defect on a cooperator? When they both defect? Pencil it on a napkin. If the napkin shows a matrix where defection always pays better than cooperation, no Game Technique will rescue it — you have shipped a Prisoner’s Dilemma and your players are doing exactly what the theory predicts.

Use CD1 (Epic Meaning & Calling) to install a coordination wrapper

The single highest-leverage move in interdependence-aware design is to wrap a competitive or mixed-motive matrix inside a CD1 narrative big enough that defection at the inner level reads as betrayal at the outer level. World of Warcraft raid mechanics are formally Stag Hunts, but the social punishment for defection in a guild-run raid comes from the outer cooperative wrapper of guild reputation. Without that wrapper, the inner matrix would devolve in weeks. Designers who try to ship guild content without the reputational scaffolding consistently watch it die.

Translate Investment Model variables to CD4 (Ownership & Possession) levers

Satisfaction maps to CD2 (Development & Accomplishment) feedback loops. Alternatives map to CD6 (Scarcity & Impatience) and CD8 (Loss & Avoidance) restrictions on outside options. Investment maps directly to CD4 (Ownership & Possession) accumulation mechanics — the points, badges, custom avatars, social graphs, and unlocked content that a user couldn’t take with them if they left. Most retention designers understand the third lever; the strongest retention features in the past decade pull on all three at once.

Distinguish exchange and communal Game Techniques

Some Game Techniques work in exchange matrices and break in communal ones, and vice versa. Direct gifting (a CD5 staple) works beautifully in communal contexts (think the gift economy in MMOs between friends) and reads as bribery in exchange contexts (a vendor who gives a “free” demo). Mentorship works in communal arrangements where the mentor isn’t tracking ledger; it falls apart the moment the mentor expects strict reciprocity. Designers who plug Game Techniques into the wrong matrix type get the right Core Drive activated against the wrong relationship grammar — and the resulting feeling is creepy, not warm.

Use CD3 (Empowerment of Creativity & Feedback) to widen the strategy space

Mixed-motive matrices stay interesting forever when players can invent new moves, new alliances, new ways to cooperate or defect. Diplomacy is fascinating because the strategy space is huge. Tic-tac-toe is fascinating once. CD3 mechanics — user-generated content, emergent strategy, social deduction — expand the matrix from 2×2 to 2×n, and that explosion is what keeps the social interaction generative.

Use CD7 (Unpredictability & Curiosity) to keep the partner’s move uncertain

Interdependence theory’s “information availability” dimension is the lever you turn when you want a social interaction to feel high-stakes. Reveal both moves simultaneously and the matrix is a one-shot game. Hide one player’s move and the other is forced to predict, model, and trust. Social-deduction games (Werewolf, Among Us) are masterclasses in calibrated information asymmetry. Most cooperative apps over-share information for transparency reasons and accidentally drain the relational tension that made the feature compelling in the first place.

Practical Steps to Apply Interdependence Theory

Concrete moves you can run on a feature spec next week.

Step 1: Draw the matrix. Pick the two main choices each player has in the feature. Sketch the four-cell matrix. Fill in payoffs honestly — including hidden time costs, reputation costs, opportunity costs. If you can’t fill in the cells, the feature’s incentive structure isn’t clear enough yet, and that ambiguity is what your users will be confused about too.

Step 2: Name the matrix. Is it a Prisoner’s Dilemma (defection always pays best individually but joint cooperation pays best collectively)? A Stag Hunt (cooperation pays best when both choose it but defection is safer)? A Chicken (mutual defection is catastrophic, mutual cooperation is fine, and one-sided defection wins)? A Coordination Game (no defection incentive at all, just a matching problem)? Different matrices need different design fixes. You can’t fix what you can’t name.

Step 3: Audit the six dimensions. Walk through degree (how interdependent are the players, really), mutuality (do they depend on each other equally), correspondence (do their interests align or oppose), basis (exchange or coordination), temporal structure (one-shot or repeated), and information availability (who knows what the other will do). Each dimension is a design lever. Most CD5 features ship with the same defaults — high mutuality, repeated, full information — and play it safe in ways that flatten the social experience.

Step 4: Identify the transformation rule you want users to apply. Self-interest? Equity? Joint maximization? Partner-protection? Each transformation produces different effective matrices and different equilibria. The design question is which transformation your reward and feedback architecture trains users to apply over time. The first 30 days of a multiplayer feature’s life are the period when users are learning which transformation rule the system rewards. Most products fail this teaching unintentionally.

Step 5: Install the cooperation wrapper. If your inner matrix has any competitive or mixed-motive elements, you need a higher-level cooperative wrapper that punishes defection at the meta level. Reputation systems, community moderation, guild membership, public profiles, identity continuity — these are the wrapper. Without one, the local-rational defection in the inner matrix will eventually swamp any cooperative norm.

Step 6: Tune information availability. If the matrix is feeling too predictable (boring), add information asymmetry. If it’s feeling too random (frustrating), remove some. The dimension that most designers ignore is also the one with the most range.

Step 7: Measure commitment, not satisfaction. If you have a long-term retention goal, instrument the three Investment Model variables (satisfaction, alternatives, investment) and look at their interaction, not just satisfaction alone. A user with falling satisfaction but rising investment is a different retention picture than a user with rising satisfaction but a competitor on the horizon. Most product analytics dashboards track only the first variable, and they miss the actual leading indicator.

Step 8: Stage the matrix changes. A new feature that suddenly shifts an established matrix shape — from coordination to mixed-motive, say, by adding a competitive ladder to a previously cooperative product — will be experienced by users as a betrayal even if the new structure is, on balance, healthier. The transformation rule users have been applying was tuned to the old matrix; switching to the new one mid-game without warning forces them to reassess every relationship the product was hosting. Stage the change. Pre-announce it. Run it as an opt-in for a generation of users before flipping the default. This is the lesson of every PvP-mode launch in a previously PvE-only game that goes badly.

Step 9: Audit your moderation as wrapper infrastructure, not policy. If you have a community feature with any meaningful interdependence, your moderation tools are not a content-policy concern — they are the cooperation wrapper that lets the inner matrix sustain itself. Underfunded moderation is the structural reason mixed-motive matrices collapse into pure defection over time. The most experienced community designers I’ve worked with have a saying: budget for moderators the way you budget for servers. The matrix is the servers; the moderators are the cooling system; without one the other catches fire.

Closing Thoughts

The reason I keep recommending interdependence theory to designers who feel they already know cooperation is that it gives you a way to be specific about the structural shape of a social interaction without resorting to platitudes about “community” or “belonging.” Once you can sketch the matrix, you can argue about it. You can say, with precision, “this guild feature is shaped like a Prisoner’s Dilemma and that’s why our healers feel exploited.” You can say, “the leaderboard is shipping in an exchange matrix and we’re asking for communal behavior.” Those statements are testable, refutable, and actionable in a way that “our community is toxic” never is.

The theory has limits — the dyadic matrix is a poor model of network dynamics, the transformation rule remains under-specified, and the Investment Model leans hard on self-report — but the limits are knowable, and a designer who knows them is in a better epistemic position than one working from intuition alone. Octalysis tells you what to engineer for. Interdependence theory tells you which structural arrangement makes the engineering possible. Combined, they are most of the toolkit you need to ship multiplayer features that don’t collapse into the same predictable failure modes.

If you only take one thing from this guide, take this: the next time a multiplayer feature underperforms, your first move shouldn’t be to add more rewards or more reminders. It should be to draw the matrix.

Frequently Asked Questions

What is interdependence theory in one sentence?

Interdependence theory is a framework, developed by Harold Kelley and John Thibaut starting in 1959 and formalized in 1978, that analyzes social interactions as outcome matrices in which each person’s payoffs depend on the choices of both themselves and their partner.

Who created interdependence theory?

Harold H. Kelley and John W. Thibaut. Their 1959 book The Social Psychology of Groups introduced the matrix vocabulary. Their 1978 follow-up Interpersonal Relations: A Theory of Interdependence formalized the six structural dimensions and the distinction between the given and effective matrix that anchors modern applications.

What is the difference between the given matrix and the effective matrix?

The given matrix is what the situation literally pays out for each combination of choices. The effective matrix is the matrix the person actually decides on after applying a transformation that re-weights outcomes for the partner’s welfare, the relationship’s long-term value, and personal values. People rarely play the given matrix straight — they play the matrix as transformed.

What are the six structural dimensions of interdependence?

Degree of dependence (how much your outcomes vary with the partner’s choice), mutuality (do you depend on each other equally), correspondence vs conflict of interest (do your payoffs align), basis of dependence (exchange vs coordination), temporal structure (one-shot, sequential, or extended), and information availability (do you know each other’s payoffs and moves). Together, they classify the structural shape of any interaction.

What is the Investment Model and how does it relate to interdependence theory?

Caryl Rusbult’s 1980 Investment Model is the most empirically tested descendant of interdependence theory. It says relationship commitment is a function of three variables: satisfaction with the relationship, quality of available alternatives, and investment already sunk into the relationship. Commitment in turn predicts persistence behaviors like accommodation, sacrifice, and forgiveness.

What are exchange relationships vs communal relationships?

Margaret Clark and Judson Mills distinguished exchange relationships, in which benefits are tracked and reciprocated like a ledger, from communal relationships, in which benefits are given in response to need or to please the partner. The two types follow different rules; importing exchange logic into a communal relationship reads as cold, and importing communal logic into an exchange relationship reads as exploitative.

How does interdependence theory differ from game theory?

Game theory and interdependence theory share the matrix vocabulary but answer different questions. Game theory is prescriptive — what should a rational actor do given a matrix? Interdependence theory is descriptive and psychological — what will a real person, with messy other-oriented preferences, actually do? The transformation from given to effective matrix is the part interdependence theory adds and game theory does not.

How does Social Value Orientation fit in?

Paul Van Lange’s Social Value Orientation (SVO) measures a person’s stable disposition toward others’ outcomes — prosocial (jointly maximize), individualistic (maximize self), or competitive (maximize relative to other). SVO predicts which transformation rule a person tends to apply to interdependent situations, partially answering interdependence theory’s open question of which transformation will be reached for in a given case.

What are the strongest critiques of interdependence theory?

Three critiques recur. The dyadic 2×2 matrix doesn’t scale cleanly to triads or networks. The transformation rule remains a black box — we can describe it post hoc but rarely predict it in advance. And the Investment Model’s heavy reliance on self-report measures inflates effect sizes relative to behavioral outcomes.

How do designers use interdependence theory?

Designers use interdependence theory to diagnose the structural shape of any multiplayer or social feature before deciding which Octalysis Game Technique to apply. The matrix tells you whether you’re shipping a Prisoner’s Dilemma, a Stag Hunt, a coordination game, or a mixed-motive game. Different matrix shapes need different cooperation wrappers, information-asymmetry tunings, and reward architectures — and the wrong design move on the wrong matrix shape consistently destroys the trust the feature was meant to build.

References

1. Thibaut, J. W., & Kelley, H. H. (1959). The Social Psychology of Groups. New York: Wiley.
2. Kelley, H. H., & Thibaut, J. W. (1978). Interpersonal Relations: A Theory of Interdependence. New York: Wiley.
3. Kelley, H. H., Holmes, J. G., Kerr, N. L., Reis, H. T., Rusbult, C. E., & Van Lange, P. A. M. (2003). An Atlas of Interpersonal Situations. Cambridge: Cambridge University Press.
4. Rusbult, C. E. (1980). Commitment and satisfaction in romantic associations: A test of the Investment Model. Journal of Experimental Social Psychology, 16(2), 172-186.
5. Rusbult, C. E., & Van Lange, P. A. M. (2003). Interdependence, interaction, and relationships. Annual Review of Psychology, 54, 351-375.
6. Rusbult, C. E., Verette, J., Whitney, G. A., Slovik, L. F., & Lipkus, I. (1991). Accommodation processes in close relationships: Theory and preliminary empirical evidence. Journal of Personality and Social Psychology, 60(1), 53-78.
7. Van Lange, P. A. M. (1999). The pursuit of joint outcomes and equality in outcomes: An integrative model of social value orientation. Journal of Personality and Social Psychology, 77(2), 337-349.
8. Clark, M. S., & Mills, J. (1979). Interpersonal attraction in exchange and communal relationships. Journal of Personality and Social Psychology, 37(1), 12-24.
9. Clark, M. S., & Mills, J. (1993). The difference between communal and exchange relationships: What it is and is not. Personality and Social Psychology Bulletin, 19(6), 684-691.
10. Le, B., & Agnew, C. R. (2003). Commitment and its theorized determinants: A meta-analysis of the Investment Model. Personal Relationships, 10(1), 37-57.
11. Le, B., Dove, N. L., Agnew, C. R., Korn, M. S., & Mutso, A. A. (2010). Predicting nonmarital romantic relationship dissolution: A meta-analytic synthesis. Personal Relationships, 17(3), 377-390.
12. Holmes, J. G. (2002). Interpersonal expectations as the building blocks of social cognition: An interdependence theory perspective. Personal Relationships, 9(1), 1-26.
13. Mischel, W., & Shoda, Y. (1995). A cognitive-affective system theory of personality: Reconceptualizing situations, dispositions, dynamics, and invariance in personality structure. Psychological Review, 102(2), 246-268.
14. Murray, S. L., Holmes, J. G., & Collins, N. L. (2006). Optimizing assurance: The risk regulation system in relationships. Psychological Bulletin, 132(5), 641-666.
15. Balliet, D., Mulder, L. B., & Van Lange, P. A. M. (2011). Reward, punishment, and cooperation: A meta-analysis. Psychological Bulletin, 137(4), 594-615.

Related Reading

• Social Identity Theory — the disposition layer that decides which transformation rule a person applies to a given matrix.
• Realistic Conflict Theory — the macro-level case for what happens when groups face genuinely opposing matrices.
• Intergroup Contact Theory — Allport’s structural conditions for prejudice reduction, all of which are matrix-shape decisions.
• Self-Determination Theory — the individual-level needs that interdependence theory’s structures must serve to feel good.
• Investment Model (Rusbult) — the most empirically tested descendant of interdependence theory, formalizing commitment as a function of satisfaction, alternatives, and investment.
• The Octalysis Framework — the eight Core Drives that interdependence-shaped features ultimately have to engage to deliver value.

WOULD YOU LIKE YU-KAI CHOU TO WORK WITH YOUR ORGANIZATION?