The Investment Model of Commitment: An S-Tier Behavioral Designer’s Guide

Caryl Rusbult once asked a question that quietly broke half the assumptions retention designers were making: why do people stay in relationships they aren’t satisfied with? She was studying romantic couples in 1980, but the math she landed on — Commitment = Satisfaction + Investments − Alternatives — turned out to be the most useful retention equation anyone has ever written down for products, jobs, gyms, MMOs, subscription services, and yes, abusive partners.

Most behavioral-design teams I’ve advised think retention is about satisfaction. Make the product better, the user stays. That’s a satisfaction model. The Investment Model says satisfaction is roughly one-third of the answer, alternatives are another third, and the variable nobody designs deliberately — investments — is the third that decides whether your churn curve looks like a leaky bucket or a moat.

This is the post I wish every Octalysis student could read before they touch a loyalty program, a streak system, or a “your data” page. Rusbult’s model is the cleanest formal account of CD4 (Ownership & Possession) and CD8 (Loss & Avoidance) ever written, with a 50-year track record of meta-analyses, six-country replications, and one ethically uncomfortable extension to abused women in shelters that we are going to walk through honestly. By the end you’ll have the equation, the four maintenance behaviors that drop out of it, the three critiques that should keep you humble, and a concrete Octalysis playbook for designing commitment that is robust without being predatory.

Speed Run Notes

• The equation is the post. Commitment = Satisfaction + Investment Size − Quality of Alternatives. The three predictors together explain ~60% of variance in commitment across 52 meta-analyzed studies (Le & Agnew 2003).
• Investments are the variable nobody designs deliberately. Time, money, custom data, shared friends, sunk emotion, identity tied to the thing — all increase commitment independently of satisfaction. Most retention teams optimize the wrong term.
• Four maintenance behaviors drop out automatically once commitment is high: accommodation, willingness to sacrifice, cognitive interdependence, and perceived superiority of the chosen option. Spot them in your data and you’re measuring commitment directly.
• The Rusbult & Martz 1995 abused-women study is the elephant. The model predicts staying in abuse with the same equation that predicts staying in marriage — an ethical mirror anyone designing retention needs to look into squarely.
• Octalysis translation: Satisfaction is your CD2/CD3 layer, Investments is pure CD4 plus CD6, Alternatives is CD8 inverted. Design all three; don’t just stack rewards.

Table of Contents

Author Credibility: Yu-kai Chou

Yu-kai Chou created the Octalysis Framework after studying gamification since 2003 — years before the term entered mainstream vocabulary. As a Human-Systems Architect & Behavioral Designer, his framework has been applied by LEGO, Microsoft, Porsche, Coca-Cola, Salesforce, and MrBeast, impacting over 1.5 Billion Users.

Chou has taught the Octalysis methodology at Harvard, Stanford, Yale, Tesla, Google, BCG, and IDEO.

His work has been cited by Harvard, Stanford, MIT, Forbes, Wall Street Journal, Wired, US Department of Energy, NIST, NSF, NCBI, US Department of Education, ClinicalTrials.gov, and Google Scholar — with 3,700+ more academic publications. Explore his books here.

The Investment Model lives in the bloodstream of every gamification consult I’ve run for the past decade, but I’ve never written it up properly until now. It’s the framework I reach for when a client says “we have a churn problem” and what they actually have is an alternatives problem with no investment moat. I’ve watched LEGO build customer-investment depth that competitors can’t touch, advised loyalty programs where the unsung hero was custom-data investments and not points, and seen subscription services lose 40% of their users to a single competitor launch because the investment side of the equation was never deliberately designed. This guide pulls together what Rusbult actually proved, where her model breaks, and how Octalysis’s CD4 and CD8 levers operationalize the math without crossing the line into manipulation.

What is the Investment Model of Commitment

The Investment Model of Commitment is Caryl Rusbult’s 1980 extension of Interdependence Theory that explains why people stay in relationships (romantic, occupational, parasocial, or commercial) using a three-term equation. It rose out of a question Kelley and Thibaut’s 1959 / 1978 framework couldn’t answer cleanly: if interaction outcomes drive satisfaction, why do plenty of dissatisfied people refuse to leave their partners, jobs, or churches?

Rusbult’s answer was that satisfaction is one input but not the only one. Two other forces independently push people to persist. The first is what other options look like — the quality of alternatives, evaluated against a comparison standard for what a person believes they could realistically obtain elsewhere. The second is what they would lose by leaving — the investment size, the cumulative resources they have already poured into the relationship that don’t transfer out. These three terms combine into the model’s central equation:

Commitment = Satisfaction + Investment Size − Quality of Alternatives

Commitment in Rusbult’s terminology is not a feeling. It is a stable orientation toward the relationship that has three components: a long-term temporal horizon (“I expect this to last”), psychological attachment (“we” rather than “I”), and intent to persist (“I will stay through hard patches”). Commitment is what the equation produces; it is also what predicts the four maintenance behaviors covered later in this post.

The original 1980 paper studied 17 college dating couples over seven months. The 1983 follow-up tested the model on 34 couples over a year and replicated the predictive structure cleanly. The Investment Model Scale (IMS) that consultants and researchers now use was published by Rusbult, Martz, and Agnew in Personal Relationships in 1998 with strong psychometric validation across romantic, friendship, occupational, and athletic domains. By the late 1990s the model was being applied beyond dating (jobs, brand loyalty, sports-team allegiance, religious affiliation, gym memberships, and abusive relationships) with the same equation surviving across context after context.

The three predictors, in plain English

Satisfaction Level is not happiness in the moment. It is the perceived gap between what the relationship is actually delivering and what the person had calibrated themselves to expect — the comparison level (CL) Kelley and Thibaut imported from Thibaut and Kelley 1959. A person with a low CL can feel satisfied by an objectively mediocre relationship; a person with a high CL can feel dissatisfied by an objectively excellent one. Satisfaction is rewards minus costs minus CL.

Quality of Alternatives is the perceived best alternative outcome (another partner, another employer, another platform, or simply being alone) evaluated against a different standard, the comparison level for alternatives (CLalt). Alternatives include not only direct substitutes but also the option of solitude or non-membership. A person who believes nobody else would have them, that the job market is closed, or that no other product solves their need, has low alternatives quality — and high commitment, all else equal.

Investment Size is the resources tied to the relationship that would be lost or diminished if it ended. Investments fall into two buckets. Intrinsic investments are resources put directly into the relationship: time, emotional energy, self-disclosed information, custom configurations. Extrinsic investments are external resources that became entangled with the relationship: shared friends, joint property, mutual social networks, custom data, identity. The defining feature of an investment is that it does not transport — leaving incurs the loss.

The three predictors don’t merely add up. They interact. High alternatives can suppress commitment even when satisfaction is high if investments are low. High investments can sustain commitment when satisfaction is low and alternatives are high — the unhappy-but-staying pattern Rusbult set out to explain. The equation is empirical; the relative weight of each term is the variable that consultants, researchers, and product teams care about.

The Core Findings

Forty-five years of research have settled the model’s central claims with unusual clarity for a field that has lately been roiled by replication failures. The headline finding is the size of the variance the three predictors explain together. Le and Agnew’s 2003 meta-analysis aggregated 52 studies (N > 11,000) and found that the three terms jointly accounted for roughly 60% of the variance in commitment scores — an effect size that puts the Investment Model in the top tier of social-psychology models for predictive power, alongside the Theory of Planned Behavior.

The individual term-level effect sizes from Le and Agnew were Satisfaction r = 0.68, Investment Size r = 0.46, and Alternatives r = −0.48 (the negative sign meaning that better alternatives lower commitment). Each term carried an independent contribution above and beyond the others — that is, the three predictors are not redundant. A second meta-analysis by Etcheverry and Le in 2005 added longitudinal data and confirmed the predictors prospectively forecast breakup at 3-, 6-, and 12-month windows.

The four pro-relationship maintenance behaviors

The most useful operational consequence of the model is its prediction of four downstream behaviors that follow from high commitment. Rusbult’s collaborators (Wieselquist, Foster, and Agnew) catalogued these across the 1990s, and they map directly onto behavioral signals retention designers can observe in product data.

Accommodation is the willingness to inhibit a destructive impulse and respond constructively when the partner behaves badly. In couples this looks like staying calm when criticized; in products it looks like a user who reports a bug rather than churning when something breaks. Highly committed users send tickets, write feedback, and engage with apologies; uncommitted users walk away silently.

Willingness to sacrifice is forgoing personally desired outcomes for the relationship’s benefit. In product terms it looks like users who tolerate a price hike, a feature deprecation, a slower experience, or a temporary outage rather than switching. Sacrifice is asymmetric (a measurable behavior, not an attitude), and the willingness to do it under pressure is one of the cleanest validations that commitment is high.

Cognitive interdependence is the gradual blurring of self and other in mental representation. In couples, the partner’s outcomes start to be experienced as one’s own. In products, this is the moment a user’s identity starts being expressed through the platform (“my Spotify Wrapped,” “my Strava,” “my Notion”) rather than the platform being experienced as an external tool. Linguistic markers (“we,” “us,” “ours”) in user research transcripts and reviews are the cheapest measurement.

Perceived superiority of partners is the cognitive distortion that makes the chosen option look better than alternatives even after fair comparison. Highly committed users systematically rate competitors lower than blind benchmarks would predict. It’s a derogation effect that protects the commitment, and it’s why straight feature comparisons against competitors rarely move loyal users.

Wieselquist, Rusbult, Foster, and Agnew’s 1999 mutual cyclical growth model added one more layer: high commitment increases trust, trust increases pro-relationship behavior, pro-relationship behavior is reciprocated, reciprocation increases satisfaction, and satisfaction feeds back into commitment. The model is not just static prediction; it describes a positive feedback loop that compounds over time.

Cross-domain replication

The Investment Model has now been validated across romantic relationships in heterosexual and same-sex samples, friendships, family relationships, employees and their employers, athletes and their teams, churchgoers and their congregations, gym members, college students and their universities, and consumers and their brands. The Le and Agnew meta-analysis spans North America, Europe, Israel, and East Asia. Sprecher, Sullivan, and Hatfield 1994 replicated in U.S. and Russian samples; Lin and Rusbult 1995 replicated in Taiwan; Van Lange, Rusbult, Drigotas, Arriaga, Witcher, and Cox 1997 replicated across Dutch, American, and German samples. The cross-cultural pattern is not perfect (we’ll get to where it isn’t), but the model travels better than most.

What Rusbult Got Right

The standard critique of social psychology is that it produces clever experiments that don’t generalize. Rusbult did the opposite. She started with a real-world puzzle — staying in unsatisfying relationships — built a math-friendly model that could be misapplied if she got it wrong, and then spent thirty years gathering the longitudinal evidence that confirms or refutes the equation. That methodological discipline is most of what’s right about the model.

The first thing she got right was decoupling commitment from satisfaction. Before 1980 the dominant assumption in relationship research was that staying tracked liking. Rusbult’s data showed dissatisfied people staying for predictable reasons and satisfied people leaving for predictable reasons, and the equation explained both. For retention designers this is the single most important shift in the model: stop measuring satisfaction and assuming it predicts retention. Measure commitment, and decompose it into the three terms.

The second thing she got right was treating investments as an independent causal variable rather than a sunk-cost confusion. Most economists treat sunk costs as irrational anchors that humans should ignore (but don’t). Rusbult treated them as functionally rational: the things you’ve put into the relationship that you would lose by leaving are real costs of leaving, and any decision-theoretic account of staying needs to include them. Recent neuroeconomic work on loss aversion (see Prospect Theory) has caught up to her position.

The third thing she got right was building the Investment Model Scale with strong psychometrics and giving it away. The 1998 IMS in Personal Relationships is freely available, has been translated into more than fifteen languages, and has consistent reliability coefficients (alphas typically 0.85+) across populations. The scale’s openness is a major reason the cross-cultural replication record is as strong as it is — researchers can pick up an instrument they trust and apply it locally.

The fourth thing she got right, and it’s the one that matters most for product design, was the prediction of pro-relationship maintenance behaviors. Most retention models stop at “did the user stay?” Rusbult predicted four observable downstream behaviors, all of which can be measured passively in product data: tolerance of bugs and pricing changes (sacrifice), bug reports and feedback (accommodation), use of possessive linguistic markers (cognitive interdependence), and depressed sensitivity to competitor benchmarks (perceived superiority). A retention dashboard built on these signals is much more diagnostic than a churn-rate metric alone.

The fifth thing she got right was honest engagement with the model’s most uncomfortable implication. Rusbult and Martz’s 1995 study of women in domestic-violence shelters did not flinch from showing that the equation predicts staying in abuse the same way it predicts staying in marriage. We’ll discuss this study more in the Elephant section — but the point I want to make here is that she didn’t try to bend the model to make abuse look like an exception. She let the model speak, and used it to argue for structural alternatives-creation as a leverage point for intervention. That intellectual honesty is rare.

Where the Investment Model Falls Apart

The model has been replicated to within an inch of its life and the equation holds. That doesn’t mean it’s complete. The areas where it falls apart are not failures of replication; they are limitations of scope. Three of them matter for behavioral designers.

It treats commitment as if it were rationally calculated

The Investment Model assumes the three predictors are weighed against one another in something like a deliberate cost-benefit way. Empirically, much of what people call commitment is not deliberated — it is identity-fused. Swann’s 2009 work on identity fusion shows that some forms of allegiance (family, religion, nationality, hardcore fandom) are experienced as constitutive of the self rather than as relationships one has. People won’t run the equation on a fused commitment because there is no separate self to do the running. They die for fused groups. They don’t recompute alternatives.

For product designers this matters because the most resilient forms of consumer loyalty (Apple, Harley-Davidson, Liverpool FC, certain MMOs) have crossed the line from investment-driven commitment into identity-fused membership. The Investment Model can describe how a user moves toward fusion, but it can’t predict what happens after fusion occurs. Once a user is fused, the equation underestimates retention because investments and alternatives are no longer being separately evaluated against the self — they’re part of the self. The model degrades gracefully here, but you stop being able to use it to forecast behavior.

Investments are partly post-hoc justifications

Rusbult treated investments as causes of commitment. Cognitive Dissonance Theory raises the awkward possibility that investments are sometimes consequences of commitment instead. People who have decided to stay invent or amplify investments to justify their choice; people who have decided to leave under-report the investments that exist. The arrow of causation is not clean.

Beach and Tesser’s confirmation work and Goldsmith and Andrade’s later studies show people retroactively re-rate investment size as a function of their current decision direction. This isn’t a death blow — longitudinal designs (Etcheverry & Le 2005) do show prospective prediction — but it does mean cross-sectional IMS scores are not pure causes. They are partly the product talking to itself in the user’s head.

For designers, the implication is that survey data on investments is contaminated. The same custom playlist that a Spotify lifer rates as enormous, a Spotify churner rates as trivial, even when the data says they’re identical. If you’re trying to model retention, behavioral measures of investment (literal counts of custom data, hours logged, friends added) outperform self-report.

It under-measures relational forms outside Western individualism

The cross-cultural replication record is often cited as a strength. Look closer and the cracks appear. In collectivist samples (China, India, Latin America), the model still predicts commitment but the relative weights of the three predictors shift markedly. Investments dominate in collectivist samples; satisfaction dominates in individualist samples; alternatives matter more or less depending on whether the relationship is publicly recognized. This is the Hofstede-dimension shift you’d predict, but it puts a ceiling on how universally the equation applies with constant coefficients.

Rusbult also implicitly assumed the three predictors are dyadic — properties of a single relationship. In cultures where relationships are more triadic or networked (arranged marriages embedded in extended family obligations, work loyalties tied to multi-generational firms, religious commitments mediated by communal accountability), the unit of analysis is wrong. Adams and Plaut’s 2003 work on relational mobility, and more recently Thomson, Yuki, and others, show that the very concept of “alternatives” is differently structured in low-relational-mobility societies. The IMS measures something there, but not exactly the same thing it measures in a U.S. dating sample.

For product teams designing for global audiences, the practical lesson is that the satisfaction / investment / alternatives lever ratios that worked in U.S. user research won’t transfer cleanly. Re-test the relative weights locally before assuming a retention strategy generalizes.

The Brain on the Investment Model

The neuroscience of commitment is younger than the model itself, but the picture that has emerged maps cleanly onto Rusbult’s three predictors. Each term lives in roughly identifiable neural circuits, and the prediction-error signals that update commitment over time look like a textbook case of reinforcement learning.

Satisfaction tracks reward prediction errors in dopaminergic midbrain regions (ventral tegmental area, nucleus accumbens). When outcomes exceed the comparison level, dopamine bursts; when they fall short, it dips below baseline. The classic Schultz primate work and the human fMRI replication by O’Doherty and colleagues established this structure. The implication for the Investment Model is that satisfaction is not a stable trait — it is an integrated stream of small prediction errors against the user’s calibrated expectation. This is why “delight” interventions that exceed expectations have outsized commitment effects relative to their cost.

Investments live in a different circuit. Loss anticipation activates the anterior insula, the amygdala, and the right inferior frontal cortex — the same loss-aversion circuitry that Prospect Theory rides on. Knutson and colleagues’ fMRI work shows the anterior insula activates when participants contemplate losing accumulated tokens roughly twice as strongly as the accumbens activates when they contemplate equivalent gains. That 2:1 asymmetry is the neural basis of why investments are weighted more heavily than equivalent rewards in the commitment equation. It is also why removing accumulated user state (a streak, a custom configuration, a friend list) produces churn out of all proportion to its rational value.

Alternatives engage a comparison circuit centered on the medial prefrontal cortex and ventral striatum. When participants evaluate a current option against an alternative, the mPFC encodes a relative-value signal that subtracts alternative from current. Rangel and Padoa-Schioppa’s value-coding work suggests this is the same circuit used in any comparative choice. The implication is that alternatives have to be made salient in working memory to influence the equation. A platform that doesn’t surface competitors in the user’s mental model has, by neural mechanism alone, depressed the alternatives term.

The Wieselquist mutual cyclical growth model maps onto the brain’s positive-feedback loops between dopamine reward signaling and oxytocin-mediated affiliation. High commitment correlates with elevated baseline oxytocin (Schneiderman, Zagoory-Sharon, Leckman, & Feldman 2012), which lowers the threshold for trust-extending behavior, which produces reciprocal trust extensions, which feed back into more frequent reward signaling. The full model is not yet at fMRI replication maturity, but each link in the chain has individual support.

Investment Model vs Other Theories

The Investment Model is most often confused with four nearby frameworks. Distinguishing them sharpens the design moves each one suggests.

Framework	What it claims	Where it differs from the Investment Model	Design move it suggests
Investment Model (Rusbult 1980)	Commitment = Satisfaction + Investments − Alternatives	—	Design all three terms; treat investments as a separately tunable retention lever.
Sunk-Cost Fallacy	Past investments irrationally bias future decisions.	Treats investment-weighting as a bug, not a rational input. The line between rational investment and sunk cost is whether the user weights at or above replacement cost.	Audit whether the user’s investment is being weighted above its real replacement cost — if so, you are inducing sunk-cost bias, not durable commitment.
Endowment Effect	People inflate the value of what they already possess.	Narrower scope: ownership only. Investment Model adds time, identity, and social-graph entanglement.	Don’t confuse giving rewards (endowment) with cultivating user-generated state (investment).
Self-Determination Theory (Ryan & Deci)	Satisfaction = autonomy + competence + relatedness.	Composes. SDT decomposes the satisfaction term; Investment Model still owns investments and alternatives.	Use SDT to diagnose why satisfaction is low, then plug it back into the IMS equation.
Interdependence Theory (Kelley & Thibaut)	Parent framework with comparison level (CL) and CLalt.	Investment Model is the special case that adds the investments term and tests it empirically.	Read Interdependence Theory for cooperation/competition structure; read this post for retention specifically.

Investment Model vs Sunk-Cost Fallacy. The sunk-cost fallacy treats accumulated investments as an irrational anchor that should be ignored. The Investment Model treats them as a rational input to staying because their loss is real. Both descriptions can coexist. People sometimes weight investments above their actual replacement value (irrational sunk cost), and they sometimes weight them at their actual replacement value (rational investment). The empirical question is whether the user’s investment weighting tracks the actual cost of replacement. If it does, you are activating the Investment Model. If it exceeds it, you are activating the sunk-cost fallacy and you should be uncomfortable. See Sunk Cost Fallacy for the line.

Investment Model vs Endowment Effect. The endowment effect is the inflated valuation of objects merely because the person possesses them. It overlaps with investments but is narrower — endowment is about ownership, investments are about entanglement. A custom playlist invests me in Spotify because removing it incurs a real loss; a generic gift card from Spotify endows me with $5 of buying power but doesn’t invest me in the platform. Designers conflate the two and end up giving away rewards (endowment) when they should be cultivating user-generated state (investment).

Investment Model vs Self-Determination Theory. SDT says satisfaction in any relationship is a function of three psychological nutrients: autonomy, competence, and relatedness. The Investment Model treats satisfaction as a single term but is silent on its composition. The two frameworks compose — you can use SDT to decompose satisfaction into the three nutrients, then plug it into the Investment Model alongside investments and alternatives. Combining the two is what most well-functioning loyalty programs accidentally do.

Investment Model vs Interdependence Theory. Interdependence Theory is the parent framework. It supplies the comparison-level concept (CL and CLalt) and the outcome-matrix machinery; the Investment Model is the special case that adds the investments term. Designers who care about the dyadic structure of cooperation and competition should read Interdependence Theory; designers who care about retention specifically should read this one.

The Investment Model in the Real World

The Investment Model has been applied across more contexts than any retention model I know. The four below are the ones I lean on most often when advising clients, and they capture the distinct ways the equation rebalances across domains.

Romantic relationships and the original data

Most of the empirical record is here, so it’s the cleanest baseline. In Le and Agnew’s 52-study meta-analysis of dating and married couples, the three predictors carry weights that line up with intuition: satisfaction is the largest beta (~0.50), investments and alternatives are near-equal contributors (~0.30 each, opposite signs). Couples who report high satisfaction, high investments, and low alternatives are the most committed and the least likely to break up at 12-month follow-up. The four maintenance behaviors are observable in their interaction patterns: they accommodate, sacrifice, talk in “we” terms, and rate other potential partners as worse than blind raters.

The under-discussed finding is that satisfaction can drop dramatically without breakup if investments are high. Bui, Peplau, and Hill’s 1996 fifteen-year follow-up of dating couples showed that 30% of long-term married pairs had crossed periods of low satisfaction without divorce because investments (joint property, shared children, joint friends, identity as a couple) were carrying the equation. This is the empirical pattern that shows up in every retention curve I’ve ever seen for a high-investment platform.

Employment and organizational commitment

Allen and Meyer’s three-component model of organizational commitment (1990) is functionally an Investment Model dialect for employer-employee relationships. Affective commitment tracks satisfaction; continuance commitment tracks investments minus alternatives (literally how Meyer and Allen define it: “perceived costs of leaving”); normative commitment is a separate moral-obligation term not in Rusbult’s equation. The Allen-Meyer scale and the IMS correlate at about r = 0.65 across employment samples. For HR designers, the operational implication is that retention bonuses that increase investments (vesting schedules, accumulated leave, relocation packages) push continuance commitment but don’t move affective commitment unless the work itself is satisfying. Pure golden-handcuffs strategies create the dissatisfied-but-staying employees that Rusbult specifically predicted — and that don’t accommodate, don’t sacrifice, and don’t perceive their employer as superior. Their accommodation absence shows up as a quiet-quit pattern.

Products, subscriptions, and brand loyalty

Bettencourt and Brown 2003 and Verhoef 2003 applied the Investment Model to consumer-brand relationships. The retention pattern that drops out is the one I see in client engagements every quarter: subscription services with low investment design lose users to competitor launches at near-1:1 rates with the competitive feature delta, while subscription services with high investment design retain users through 30-50% feature gaps. The investment design moves I track include: custom user state (playlists, wishlists, configurations), accumulated history (watch lists, orders, conversation logs), social graph entanglement (friends on the platform, shared content), platform-specific identity (handles, achievements, public profile), and custom integrations (third-party connections, automations, exports configured). When churn spikes after a competitor launch, the diagnostic question is which of those investment terms were not designed deliberately. The answer is usually all of them.

The instructive case is Spotify versus Apple Music. Apple has higher device-tier integration and broadly equivalent satisfaction. Spotify has dramatically higher investments — user-curated playlists, listening history, social-graph follows, Spotify Wrapped as identity artifact, third-party integrations. The Investment Model predicts that Spotify users, even when objectively dissatisfied, churn less than the satisfaction differential alone would predict. The data confirms it.

Gyms, MMOs, and the streak economy

Streaks are a pure Investment Model design. A streak has no satisfaction value — nobody enjoys the streak per se. Its function is to accumulate an investment whose loss-on-leaving exceeds the marginal cost of staying. Duolingo’s streak system, the Headspace consecutive-day count, MMO daily-quest chains, and the gym-class “unbroken attendance” loyalty tiers all exploit the same mechanism. The user is not staying because the experience is good; they are staying because they would lose the streak.

This is also where the model crosses an ethical line designers should be alert to. A streak that accumulates investment without satisfaction is the textbook definition of a Black Hat retention loop: the user persists not because the relationship serves them but because leaving incurs loss. The combination of high satisfaction and high investments is healthy; the substitution of investments for satisfaction is parasitic. We’ll come back to this in the Elephant.

The Elephant in the Room

The most ethically uncomfortable thing about the Investment Model is the Rusbult and Martz 1995 study of women in domestic-violence shelters. Rusbult and Martz applied the IMS to 100 women who had left abusive partners and were now in shelters. With disturbing accuracy, the model predicted which women would return to their abusers within six months: high investments and low alternatives forecasted return at strong effect sizes, even when satisfaction was unambiguously low.

Their 1995 paper concluded that the same equation that explains why people stay in marriage explains why people return to abuse. This equation is morally neutral; it describes what holds people in relationships regardless of whether the holding is good for them. Rusbult did not present this as a defense of staying. She argued that intervention strategies needed to focus on the alternatives term (building viable economic, housing, and social options outside the relationship), because satisfaction interventions (“but he’s bad for you”) couldn’t move commitment when investments were high and alternatives were perceived as worse.

The lesson for product designers, and the reason this is the elephant section, is that high commitment is not always a sign of a healthy relationship. The Investment Model is the equation behind why users stay; it is also the equation behind why users stay when they shouldn’t. A retention strategy that builds large investments while satisfaction silently degrades is producing the user-equivalent of the women in Rusbult’s shelter sample — people who are committed because leaving costs more than staying, not because what they have is good.

The line I draw with clients is this: investments are ethical when they accumulate alongside genuine satisfaction and when leaving is structurally possible (data export, identity portability, no surprise loss penalties). Investments cross into manipulation when they substitute for satisfaction, when leaving destroys disproportionate value, or when the platform deliberately suppresses the alternatives term by hiding competitors, blocking exports, or leveraging social pressure. The Investment Model gives you the levers; ethical design is what you choose to do with them.

How to Apply the Investment Model with the Octalysis Framework

The Investment Model is the cleanest behavioral-science backbone for the Octalysis Framework’s retention layer that exists. Each of Rusbult’s three terms maps to a specific Octalysis Core Drive cluster, and the four maintenance behaviors map to observable Octalysis-aligned signals. Below is the canonical translation I teach.

Satisfaction → CD2 + CD3 (and the White Hat upper half)

Satisfaction is what the user gets while they are in the relationship. In Octalysis terms, it is delivered through Core Drive 2 (Development & Accomplishment) and Core Drive 3 (Empowerment of Creativity & Feedback) — the White Hat drives that produce a sense of progress, mastery, and meaningful agency. Game Techniques that deliver here include Progress Bars (#4), Step-by-Step Overlays (#6), Boss Fights (#14), and meaningful Glowing Choice mechanics (#28). If you can’t articulate which CD2 and CD3 mechanics are firing for your committed users, your satisfaction term is under-designed.

A common failure is over-relying on Core Drive 1 (Epic Meaning & Calling) for satisfaction. Mission framing is powerful but doesn’t replace the lived satisfaction of a CD2/CD3 progression loop. Mission gets users in the door; CD2/CD3 keeps the satisfaction term high.

Investments → CD4 + CD6

Investments is the design variable Octalysis names with full clarity: Core Drive 4 (Ownership & Possession) and, secondarily, Core Drive 6 (Scarcity & Impatience — the time-investment term). The Game Techniques that build investments include Collection Sets (#16), Avatars (#13), Earned Lunch (#7), Achievement Symbols (#2), and the broader Build-From-Scratch pattern (IKEA effect) where the user’s accumulated state is visible. Investment design questions to ask of any product:

• What custom state does each user accumulate that wouldn’t transfer to a competitor?
• What identity artifacts (avatar, handle, public profile, badge case) does the user build over time?
• What social-graph entanglement happens passively (mutual follows, shared content, in-product friend lists)?
• What time investments are visible to the user (streak counters, total hours, tenure markers, anniversary dates)?
• What custom configurations have they made (saved searches, filter preferences, integrations, automations)?

If you can answer all five with specific product features, you have a designed investment layer. If you can’t, you are leaning on satisfaction alone — which means you’re vulnerable to a competitor with a better satisfaction layer.

Alternatives → CD8 (inverted)

Quality of Alternatives is the term Octalysis touches via Core Drive 8 (Loss & Avoidance), inverted. CD8 in Rusbult’s frame is not “make the user fear losing what they have” but “ensure the user perceives current alternatives as inferior to staying.” Octalysis Game Techniques that depress the alternatives term include Anchored Juxtaposition, Status Quo Sloth, and the Sunk Cost Prison (#50) — the latter is technically an investment-side mechanic, but it operates by inflating the perceived loss of switching.

The ethical design question here is whether you depress alternatives by being a better product, or by making it harder for users to evaluate alternatives. The first is healthy CD8; the second is dark-pattern CD8. The line shows up in design choices like one-click data export, transparent comparison content, and willingness to publish honest reviews of competitors.

Maintenance behaviors → operationalize the four signals

The four pro-relationship maintenance behaviors are observable in product data:

• Accommodation: rate of constructive bug reports per active user, rate of feedback submissions, willingness to engage with apologies after outages.
• Sacrifice: tolerance of price hikes (churn lift after a price increase), tolerance of feature deprecation, retention through outages.
• Cognitive interdependence: linguistic analysis of user reviews, social posts, and support tickets for “we / us / our” relative to “your” or “the” markers.
• Perceived superiority: NPS asymmetry between users and matched non-users, comparative ratings of competitors over time.

Track these four as a quarterly commitment dashboard in addition to churn rate. They will identify commitment erosion months before it shows up in the churn metric, because users withdraw maintenance behaviors before they leave.

Practical Steps to Apply the Investment Model

Here is the playbook I run with clients when retention is the brief. It is not exhaustive; it is the minimum viable Investment Model audit.

Step 1: Decompose your retention metric into the three terms

Stop reporting churn as a single number. Run a quarterly survey on a representative cohort of active users using the Investment Model Scale (it’s seven items per term, twenty-one total, free to use academically and easily licensed for commercial). Score satisfaction, investment size, and quality of alternatives separately. Cross-reference with twelve-month retention. The ratio of how much each term is contributing to retention in your cohort is the diagnostic.

Step 2: Audit your investment design surface

List every form of user state in your product. For each, score: does this transfer to a competitor (yes / no), does the user perceive it as theirs (yes / no), is the loss visible if they leave (yes / no). Anything that scores three nos is a designed investment. Anything that scores three yeses is generic value that does not increase commitment. Most products I audit have one to three real investments and dozens of features they think are investments but aren’t.

Step 3: Build investment moats deliberately

The five investment categories from the Octalysis section are your design checklist: custom user state, accumulated history, social graph entanglement, platform-specific identity, and custom configurations. Pick the two weakest in your audit. Ship a feature in the next quarter that adds one investment in each category. Examples that travel well: collaborative collections, exportable-but-platform-native histories, follow graphs, branded handles, and integration libraries.

Step 4: Address the alternatives term honestly

Do not depress alternatives by hiding competitors or blocking exports. Depress alternatives by being meaningfully better in ways the user can verify. Publish honest comparison content (the perceived-superiority effect will take care of the rest once commitment is high enough). Make data exports painless. Make competitor switching visible and uncoerced. The Investment Model predicts users with portable data will still stay if your investment moat is real — and your retention strategy will be ethically defensible.

Step 5: Build the four-signal maintenance dashboard

Stand up the dashboard from the Octalysis section: accommodation rate, sacrifice tolerance, cognitive-interdependence linguistic markers, perceived-superiority asymmetry. Review it monthly. When any of the four start dropping, you have months of warning before churn moves. That window is when retention investments pay back the most.

Step 6: Run the elephant test on every retention loop you ship

For every new retention mechanic, ask: if satisfaction silently degraded by 30%, would users still feel free to leave? If the answer is no — if the design has accumulated investments to the point where leaving is structurally punishing — you have built a Black Hat retention loop, not a healthy one. Either redesign so the investments compound alongside satisfaction, or accept that you are building Rusbult’s shelter dynamic and ship with explicit ethical justification.

Closing Thoughts

Caryl Rusbult died in 2010. Her body of work is one of the cleanest examples in social psychology of starting with a real-world question and building a model rigorous enough to hold up against forty-five years of replication. The Investment Model is not a cute lab effect that doesn’t generalize; it is the equation behind why your users stay, why your employees stay, and yes, why some people stay in relationships that are hurting them.

For behavioral designers, the model is most useful as a diagnostic rather than a prescription. It tells you which term is doing the work in your retention metric, which lets you tell whether your moat is satisfaction, investments, or alternatives suppression. Each of those three has starkly different long-term implications. A satisfaction moat erodes the moment a competitor builds a better product. An investment moat compounds quietly for years. An alternatives-suppression moat is fragile, regulatorily exposed, and morally suspect.

The ethical demand the model makes of designers is that we be honest about which moat we’re building. Investments accumulated alongside satisfaction are the foundation of healthy long-term relationships between users and products. Investments accumulated as a substitute for satisfaction, or as a structural punishment for leaving, are the kind of retention I would not want my own users to feel. The Octalysis White Hat / Black Hat distinction maps directly onto the Investment Model’s ethical frontier: White Hat retention raises commitment by raising satisfaction and investments simultaneously; Black Hat retention raises commitment by raising investments and depressing alternatives while satisfaction quietly drops.

If you take one practical thing from this post, let it be this: stop treating retention as a single metric and start treating it as Rusbult’s three-term equation. Measure satisfaction, investments, and alternatives separately. Design each one deliberately. Watch the four maintenance behaviors as your early-warning signals. And every time you build a new commitment loop, run the elephant test.

If you want to put the Investment Model to work, the next step is the framework that operationalizes its three terms. Start with the Octalysis Framework pillar to see all eight Core Drives, then read the parent Interdependence Theory for the comparison-level machinery. If your team is shipping retention loops at scale and you want a teardown from the Octalysis Group, reach out here.

Frequently Asked Questions

What exactly is the Investment Model of Commitment?

It is Caryl Rusbult’s 1980 extension of Interdependence Theory, which models commitment as a three-predictor equation: Commitment = Satisfaction + Investment Size − Quality of Alternatives. The three terms together explain about 60% of the variance in commitment scores across a 52-study meta-analysis (Le & Agnew 2003). The model was originally developed for romantic relationships but has been validated across employment, brand loyalty, friendship, sports allegiance, and abusive relationships.

Who was Caryl Rusbult?

Caryl E. Rusbult (1952–2010) was an American social psychologist who is best known for her tenure at the University of North Carolina at Chapel Hill and her later position at Vrije Universiteit Amsterdam. She studied close relationships and is best known for the Investment Model and the related body of work on pro-relationship maintenance behaviors. Her research collaborators include Christopher Agnew, Stephen Drigotas, Ximena Arriaga, Caryl Wieselquist, and Paul Van Lange.

How is the Investment Model different from the sunk-cost fallacy?

The sunk-cost fallacy describes irrational over-weighting of past investments. The Investment Model describes rational weighting of investments because their loss on leaving is real. They overlap empirically but make different normative claims. A user who weights their custom playlist at exactly the cost of rebuilding it is rational by the Investment Model and free of sunk-cost bias. A user who weights it above its replacement cost is showing both Investment Model commitment and sunk-cost bias.

Does the model predict breakup as well as staying?

Yes, prospectively. Etcheverry and Le’s 2005 longitudinal meta-analysis showed the IMS predicts breakup at 3-, 6-, and 12-month windows with effect sizes around r = 0.40–0.50. Low commitment scores forecast breakup; high commitment scores forecast persistence. The longitudinal designs are part of why the model has held up to replication scrutiny.

What are the four pro-relationship maintenance behaviors?

Accommodation (constructive response to a partner’s destructive behavior), willingness to sacrifice (forgoing personal preferences for the relationship), cognitive interdependence (mental representation of self and partner as a unit), and perceived superiority of partners (rating the chosen relationship higher than blind comparisons would predict). They are downstream consequences of high commitment and serve as observable behavioral signals.

How does the Investment Model apply to abusive relationships?

Rusbult and Martz’s 1995 study of women in domestic-violence shelters found the same equation predicts return to abusive partners as predicts staying in healthy relationships. High investments and low alternatives forecasted return even when satisfaction was extremely low. The finding is ethically uncomfortable but methodologically clean. It points intervention strategies toward building real alternatives (economic, social, housing) rather than focusing on persuading victims that satisfaction is low — which they already know.

Can the Investment Model be used unethically by product designers?

Yes, and it commonly is. Investments accumulated as a substitute for satisfaction, combined with active suppression of alternatives (export blocks, hidden competitors, social-pressure dark patterns), produce the user-equivalent of Rusbult’s shelter dynamic: people who stay because leaving costs more than staying, not because what they have is good. The ethical line is whether investments compound alongside genuine satisfaction and whether structural exit remains possible.

How does the Investment Model fit with the Octalysis Framework?

Satisfaction maps to Octalysis Core Drive 2 (Development & Accomplishment) and Core Drive 3 (Empowerment of Creativity & Feedback). Investments map to Core Drive 4 (Ownership & Possession) and Core Drive 6 (Scarcity & Impatience). Quality of Alternatives maps to Core Drive 8 (Loss & Avoidance), inverted: making the user perceive alternatives as inferior. The four maintenance behaviors are observable in product data through accommodation rate, sacrifice tolerance, possessive-language frequency, and competitive-rating asymmetry.

What is the Investment Model Scale and is it free?

The Investment Model Scale (IMS) is a twenty-one item self-report instrument published by Rusbult, Martz, and Agnew in Personal Relationships in 1998. Seven items measure satisfaction, seven measure investment size, and seven measure quality of alternatives. The scale has consistent reliability (alphas typically 0.85+) across romantic, friendship, employment, and athletic samples. It is freely available for academic use; commercial licensing terms are typically permissive.

Where should I read more about the Investment Model?

Start with Rusbult’s original 1980 paper in the Journal of Experimental Social Psychology. Then read Le and Agnew 2003 in Personal Relationships for the meta-analytic synthesis, Wieselquist, Rusbult, Foster, and Agnew 1999 in Journal of Personality and Social Psychology for the mutual cyclical growth model, and Rusbult, Martz, and Agnew 1998 for the IMS. For applied/product framings, see the references list at the end of this post and the related-reading section below.

References

1. 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.
2. Rusbult, C. E. (1983). A longitudinal test of the investment model: The development (and deterioration) of satisfaction and commitment in heterosexual involvements. Journal of Personality and Social Psychology, 45(1), 101–117.
3. Rusbult, C. E., Martz, J. M., & Agnew, C. R. (1998). The Investment Model Scale: Measuring commitment level, satisfaction level, quality of alternatives, and investment size. Personal Relationships, 5(4), 357–391.
4. Rusbult, C. E., & Martz, J. M. (1995). Remaining in an abusive relationship: An investment model analysis of nonvoluntary dependence. Personality and Social Psychology Bulletin, 21(6), 558–571.
5. Le, B., & Agnew, C. R. (2003). Commitment and its theorized determinants: A meta-analysis of the Investment Model. Personal Relationships, 10(1), 37–57.
6. Etcheverry, P. E., & Le, B. (2005). Thinking about commitment: Accessibility of commitment and prediction of relationship persistence, accommodation, and willingness to sacrifice. Personal Relationships, 12(1), 103–123.
7. Wieselquist, J., Rusbult, C. E., Foster, C. A., & Agnew, C. R. (1999). Commitment, pro-relationship behavior, and trust in close relationships. Journal of Personality and Social Psychology, 77(5), 942–966.
8. Van Lange, P. A. M., Rusbult, C. E., Drigotas, S. M., Arriaga, X. B., Witcher, B. S., & Cox, C. L. (1997). Willingness to sacrifice in close relationships. Journal of Personality and Social Psychology, 72(6), 1373–1395.
9. Drigotas, S. M., & Rusbult, C. E. (1992). Should I stay or should I go? A dependence model of breakups. Journal of Personality and Social Psychology, 62(1), 62–87.
10. Lin, Y. H. W., & Rusbult, C. E. (1995). Commitment to dating relationships and cross-sex friendships in America and China. Journal of Social and Personal Relationships, 12(1), 7–26.
11. Allen, N. J., & Meyer, J. P. (1990). The measurement and antecedents of affective, continuance and normative commitment to the organization. Journal of Occupational Psychology, 63(1), 1–18.
12. Bui, K.-V. T., Peplau, L. A., & Hill, C. T. (1996). Testing the Rusbult model of relationship commitment and stability in a 15-year study of heterosexual couples. Personality and Social Psychology Bulletin, 22(12), 1244–1257.
13. Sprecher, S., Sullivan, Q., & Hatfield, E. (1994). Mate selection preferences: Gender differences examined in a national sample. Journal of Personality and Social Psychology, 66(6), 1074–1080.
14. Schneiderman, I., Zagoory-Sharon, O., Leckman, J. F., & Feldman, R. (2012). Oxytocin during the initial stages of romantic attachment: Relations to couples’ interactive reciprocity. Psychoneuroendocrinology, 37(8), 1277–1285.
15. Adams, G., & Plaut, V. C. (2003). The cultural grounding of personal relationship: Friendship in North American and West African worlds. Personal Relationships, 10(3), 333–347.

Related Reading

• Interdependence Theory — the parent framework that supplies the comparison-level concept the Investment Model rests on.
• Prospect Theory — loss aversion is the neural mechanism behind why investments are weighted asymmetrically.
• Sunk Cost Prison — Yu-kai’s Game Technique #50: the irrational sibling of the rational investments term.
• Endowment Effect — ownership-driven valuation that overlaps with but is narrower than investments.
• Self-Determination Theory — decompose the satisfaction term into autonomy, competence, and relatedness.
• The Octalysis Framework — the design system that operationalizes the model’s three terms into Core Drive levers.
• The Behavioral Framework Library — every other framework guide in this series.

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