Theory of Reasoned Action: S-Tier Behavioral Designer’s Guide

Every behavior-change theory built after 1975 reuses the same building block. You can see it in the Theory of Planned Behavior, the Health Belief Model’s revised form, the Reasoned Action Approach, the COM-B model, and almost every health-promotion campaign written in the last fifty years. The block was named, formalized, and given an equation by two researchers most designers have never read. Their model is called the Theory of Reasoned Action, and it is the most undercredited piece of intellectual property in the entire behavioral-design canon.

Most product teams treat “attitude” as a vibe. A user “likes the brand.” Customers “feel positive.” Fishbein and Ajzen rejected that framing in 1975 and replaced it with a sum. Attitude is the sum of every belief you hold about what a behavior will do for you, each weighted by how much you care about that outcome. Not a feeling. A computation. The same shape applies to social pressure: the sum of every relevant referent’s expectation of you, each weighted by how much you care what that referent thinks. The model is built out of beliefs, not constructs, which means every belief is a place a designer can intervene.

The reason this is undercredited is that the Theory of Planned Behavior, which Ajzen published in 1991 as a successor, ate the spotlight. TPB added one variable (Perceived Behavioral Control) and got a quarter-century of citations the original model deserved. But you cannot understand TPB without understanding what TRA was already doing underneath. And once you see the expectancy-value architecture, you stop trying to design “for attitudes” and start designing for the specific belief that is failing.

Speed Run Notes

• Fishbein and Ajzen 1975 isolated the variable every later behavior-change model would reuse: behavior is the output of belief-weighted attitudes plus belief-weighted norms feeding intention.
• The expectancy-value math is the lever. Attitude is not a vibe; it is the sum of (belief that the behavior produces outcome × evaluation of that outcome) across every outcome a person considers.
• The Subjective Norm side is the same shape: belief about what a specific referent expects × motivation to comply with that referent. Targets surface one belief at a time, not “norms” in aggregate.
• Compatibility Principle is why TRA field predictions fail: researchers measure attitude toward “the environment” and try to predict “recycled cans this week.” Match Action, Target, Context, Time or the model dies.
• TRA stops at intention. Sheeran 2002: 53% of intenders act. TRA names the belief layer; designers close the gap from the action layer with implementation intentions, habit, choice architecture.
• Octalysis times TRA is per-belief, not per-construct. Behavioral beliefs get Development & Accomplishment plus Ownership moves. Normative beliefs get Social Influence moves. The belief is the unit of design.

About the Creator of the Octalysis Framework

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.

TEDxLausanne · Gamification to Improve Our World17 min

What Is the Theory of Reasoned Action?

The Theory of Reasoned Action (TRA) is a model of volitional human behavior published by Martin Fishbein and Icek Ajzen in their 1975 book Belief, Attitude, Intention and Behavior. The model makes one strong claim: for behaviors a person can freely choose to do or not do, the immediate cause of the behavior is the intention to perform it, and that intention is determined by two and only two factors. The first factor is the person’s attitude toward the behavior. The second is the person’s perception of what relevant others expect them to do, which Fishbein and Ajzen called the Subjective Norm.

The contribution most readers miss is what sits underneath those two factors. Attitude is not entered into the model as a single number that the researcher assigns based on a survey question like “Do you like X?”. Attitude is computed from a list of beliefs the person holds about what the behavior will produce, each multiplied by how positively or negatively they evaluate that outcome, and then summed. The Subjective Norm has the same shape: it is computed from a list of beliefs about what specific referents expect, each multiplied by the person’s motivation to comply with that referent, then summed. This is called an expectancy-value formulation, and it traces back to Fishbein’s 1967 paper on attitude formation, which itself traces back to the expected-utility tradition in decision theory.

The whole architecture sits in five steps. Beliefs feed two intermediate constructs, the two constructs feed a third (Intention), and Intention feeds Behavior. Every arrow is a place to intervene. The boxes in the middle are not what designers should be optimizing for, even though that is what most teams do. The leverage is at the belief layer, which is the most granular layer the model exposes.

The Model in One Sentence

The shortest defensible statement of TRA: behavior is what gets done; intention predicts behavior; intention is itself predicted by a weighted combination of attitude toward the behavior and the subjective norm, where both of those are sums of belief-weighted evaluations. The weights vary by behavior and by population, which is part of why measurement specificity matters so much. Both weights are estimated empirically per study, not assumed; in the original studies the attitudinal weight was usually larger than the normative weight for personal-discretion behaviors, and the normative weight grew for more publicly visible behaviors.

Why This Was Novel in 1975

Before TRA, the dominant idea in attitude research was that attitudes directly caused behavior. If you measured a person’s attitude toward African Americans, you should be able to predict whether they would seat that person in their restaurant. The problem is that this prediction kept failing in the laboratory. The most famous demonstration was Richard LaPiere’s 1934 study, in which 250 American restaurant and hotel staff served a Chinese couple touring the United States with him almost without exception, but the same staff, when surveyed six months later by mail, said they would not serve “members of the Chinese race.” Attitude (negative) and behavior (welcoming) did not align.

For four decades after LaPiere, researchers tried to patch the attitude-behavior gap with moderator variables, situational pressure terms, and ad hoc constructs. Fishbein and Ajzen’s move was structural. They argued that the gap was a measurement artifact: researchers were measuring an attitude toward an object (Chinese people) and trying to predict a behavior (seating a specific couple in this specific restaurant). The correct measurement is attitude toward the behavior, not the object. Once you make that switch, and add subjective norms to handle the social-context part, the correlation between attitude and behavior moves from “marginal” to “strong.”

This is the move underneath. Attitudes toward objects predict almost nothing about specific behaviors. Attitudes toward behaviors, plus norms about those behaviors, predict intentions to perform them. Intentions predict actual performance. Everywhere this prediction underperforms, the methodological move is to look at the specificity of measurement before blaming the theory.

The Expectancy-Value Computation

The equation that does the actual work in TRA looks small on the page but contains the whole design surface.

Attitude_B  =  Σ (b_i × e_i)

Where b_i is the strength of the belief that performing the behavior will produce outcome i, and e_i is the evaluation of outcome i on a positive-to-negative scale. The Subjective Norm has the parallel form:

Subjective Norm_B  =  Σ (n_i × mc_i)

Where n_i is the normative belief that referent i thinks the person should perform the behavior, and mc_i is the motivation to comply with that referent.

What this means for designers is that “attitude toward the behavior” is a derived quantity, not a primary one. You do not change a customer’s attitude toward your product by asking them to like the product harder. You change it by changing one of two things: the strength of their belief that the product produces a specific outcome they value, or the value they assign to that outcome. Same for norms: you change them by changing which specific referent they consider, or the strength of that referent’s expectation, or the customer’s motivation to comply with that referent.

Why the Sum Shape Matters

The summation in those equations is doing a lot of philosophical work. It means that attitudes are compositional. They are built up out of more granular pieces, and the granular pieces can be inspected one at a time. A user whose overall attitude toward “starting a meditation habit” is mildly negative may have a strongly positive belief about the outcome “feeling calmer” and a strongly negative belief about the outcome “spending 20 minutes I do not have.” The summary attitude obscures both. The belief inventory exposes both.

For a designer, this matters because the design move is different in each case. If the negative driver is the time cost, you might lower the floor of the behavior (a 90-second guided breath instead of a 20-minute sit). If the negative driver is doubt about the outcome, you might offer evidence specific to that user’s stress profile. Both moves change the same summary attitude, but they require knowing which belief to target. The summary number does not tell you. The belief inventory does.

Behavioral Beliefs Are Not the Same as Outcome Beliefs

A common confusion in applied TRA work is the conflation of behavioral beliefs (b_i) with the user’s general beliefs about the world. A behavioral belief is specifically about what THIS behavior will produce. “Meditation reduces cortisol” is a general belief. “If I meditate for 10 minutes tonight, my anxiety in tomorrow’s meeting will go down” is a behavioral belief. Only the second one enters the TRA calculation for whether the user will meditate tonight.

This is why fact-based persuasion about general truths often fails to move actual behavior. A user can believe that “exercise prevents diabetes” with full conviction and still not exercise this week, because the behavioral belief that “going to the gym tonight will reduce my diabetes risk” is operating at a different scale of immediacy. The general belief is correct but causally distant. The behavioral belief is the one TRA says you have to move.

The Compatibility Principle (TACT)

The single most-cited methodological contribution from the TRA tradition is not the equation. It is the Principle of Compatibility, sometimes called the Principle of Correspondence, which Ajzen and Fishbein published in 1977 in Psychological Bulletin. The principle states that attitude measures and behavior measures must match on four elements: Action, Target, Context, and Time. Together these elements form the acronym TACT.

Here is what that means in practice. Suppose you are trying to predict whether college students will donate blood at a campus drive next Tuesday. The behavior, fully specified, is:

• Action: donating one unit of whole blood
• Target: to the campus blood drive operator
• Context: on this specific campus
• Time: next Tuesday between 10 AM and 4 PM

For the model to predict this behavior, the attitude question must measure attitude toward THAT exact action-target-context-time combination. “Do you have a positive attitude toward blood donation?” is not measuring the right thing. It is measuring a generalized affect toward a category. The version that actually predicts the behavior is “Donating one unit of blood at the campus blood drive next Tuesday would be . . . ?” on a good-bad scale.

This is the methodological point most behavior-change research violates. A team measures attitudes toward “the environment” or “diversity” or “healthy eating” and then wonders why those attitudes do not predict specific environmental, diversity, or eating behaviors. The model does not predict the disconnect. The disconnect is forced by the mismatch between measurement specificities. The TRA equation requires that the attitude and the behavior live at the same level of specificity. Both can be general, both can be specific, but they cannot mix.

Why TACT Matters for Design

Designers building behavior-change products absorb this principle through experience even when they have never read Ajzen and Fishbein. The reason CTAs (Calls to Action) that say “Start your free trial now, downloadable in 30 seconds” outperform “Sign up to get healthier” is that the first one matches the user’s mental model of the immediate, specific behavior they are deciding about. The second one asks the user to import a general value commitment into a specific decision moment, which is a longer cognitive trip.

The other place TACT matters is in measuring whether a design intervention worked. If you ran a campaign aimed at “increasing customer love for the brand” and then measured purchase behavior, you ran an incompatible measurement, and your null result tells you nothing. The corresponding attitudinal measure for “purchases this quarter” is “how I feel about buying this product, from this brand, in this market, during this purchase window.” That is the attitude that maps to that behavior. Anything more general is correlated but not causally close.

From TRA to TPB: Why Ajzen Added Perceived Behavioral Control

The Theory of Reasoned Action made one explicit boundary claim that Ajzen later softened. TRA was designed only for behaviors that are under volitional control, meaning behaviors the person can simply decide to do or not do. Voting. Donating blood. Buying a specific brand of detergent. For those behaviors, intention is a near-complete predictor of behavior, assuming measurement compatibility is honored.

The problem is that almost every interesting behavior is not fully under volitional control. Quitting smoking, losing weight, learning a language, completing a degree, sticking to a savings plan, getting screened for a chronic condition. These behaviors require capability, opportunity, and resources beyond bare intention. A person can fully intend to lose 20 pounds and not lose 20 pounds, not because their intention was weak but because their environment, biology, social pressure, or skill set blocked execution.

In 1985 Ajzen published From Intentions to Actions: A Theory of Planned Behavior, an extension of TRA that added a third variable: Perceived Behavioral Control (PBC). PBC captures the person’s belief about how easy or difficult performing the behavior will be, given the resources and obstacles they perceive. PBC operates two ways in the model. It feeds intention directly (people form weaker intentions to do behaviors they expect to fail at), and it also feeds behavior directly (independent of intention, if a person genuinely lacks the resources, they will not perform the behavior no matter how strongly they intend to).

This single addition is what made TPB applicable to health behavior, education, occupational behavior, and almost everything researchers actually wanted to predict. Armitage and Conner’s 2001 meta-analysis of 185 studies found that TPB accounted for 39% of variance in intentions and 27% of variance in behavior, where TRA had typically explained around 32% of intention variance and 20% of behavior variance for the same kinds of behaviors. The PBC addition was small in the diagram and large in predictive power.

What TRA Lost and What It Kept

The temptation is to read this history as TRA being superseded by TPB. That is not quite right. TRA kept doing what it was designed to do: predict volitional behavior. The expectancy-value architecture, the compatibility principle, the belief-decomposition method, the survey methodology in Predicting and Changing Behavior, all carried forward into TPB unchanged. What changed is that TPB acknowledged a class of behaviors TRA had explicitly excluded. The criticism that TRA “failed at health behavior” is really a criticism that researchers applied it past its stated boundary condition, not a flaw in the model on its own terms.

For designers, the practical implication is that you should use TRA when your behavior is genuinely volitional (clicking a link, choosing a brand, voting on something) and TPB when it is not (sustained habit, chronic-disease management, weight loss). Using TRA for sustained habits will give you the false impression that “intention is enough”, which it never is for non-volitional behaviors.

The Reasoned Action Approach (2010 Update)

In 2010 Fishbein and Ajzen published Predicting and Changing Behavior: The Reasoned Action Approach, the integrated successor to both TRA and TPB. The Reasoned Action Approach (RAA) is what most current researchers actually use when they describe themselves as working in the Fishbein-Ajzen tradition. The book retains the equation, retains the compatibility principle, retains PBC as a third construct, and adds two important refinements.

Refinement 1: Subjective Norm splits into Injunctive and Descriptive components. The original Subjective Norm in TRA captured “what referents expect me to do” (injunctive norm: what they think I should do). RAA explicitly separates this from descriptive norm: “what referents are actually doing.” These two operate differently in different contexts. Injunctive norm dominates when the referent has authority or evaluative power (a doctor, a parent, an institutional review board). Descriptive norm dominates when the referent is a peer group the person identifies with and is using to gauge what is normal. Most field studies before 2010 had been measuring only injunctive norms and missing half the signal.

Refinement 2: PBC splits into Capacity and Autonomy. The original Perceived Behavioral Control in TPB had been doing two jobs that RAA breaks apart. Capacity is Self-Efficacy in the Bandura sense: do I believe I can execute this behavior if I tried? Autonomy is the perception that the behavior is under my control rather than imposed or contingent on factors outside me. These two are correlated but separable; a person can feel capable of doing something while also feeling that the system would not let them. The split lets designers diagnose more precisely.

RAA in the Field

The Reasoned Action Approach is what underlies most of the public-health behavior-change literature published between 2010 and the present. McEachan and colleagues’ 2011 meta-analysis of 237 studies showed that the RAA-shape model accounted for 43% of intention variance and 23% of behavior variance for health-related behaviors, with the four-construct split (Injunctive, Descriptive, Capacity, Autonomy) outperforming the original three-construct TPB by a small but consistent margin.

For practical purposes, what this gives you as a designer is a four-belief-layer map of any behavior:

1. Behavioral beliefs (the b_i × e_i sum) feeding Attitude
2. Injunctive normative beliefs (what authority figures expect) feeding part of Subjective Norm
3. Descriptive normative beliefs (what peers are doing) feeding the other part of Subjective Norm
4. Control beliefs (split into Capacity and Autonomy) feeding PBC

Each of those four layers is its own design surface. The intervention that fixes a Descriptive-norm deficit (showing peer adoption data) is different from the intervention that fixes a Capacity-belief deficit (giving low-stakes mastery practice). The reason most behavior-change campaigns fail is that they pick one intervention and apply it to whatever belief was actually failing. The model lets you diagnose first.

What Fishbein and Ajzen Got Right

It is fashionable to relegate older models to “historical interest only,” and there is some of that energy around TRA among researchers who prefer COM-B, the Behaviour Change Wheel, or any of the dual-process derivatives. The credit Fishbein and Ajzen deserve is structural. Three things they got right have outlasted the specific findings of their 1975 book.

First: the move from attitude-toward-object to attitude-toward-behavior. This is one of the cleanest methodological corrections in the history of behavioral science. Forty years of attitude research before 1975 had assumed that liking a thing predicted doing a thing about the thing, and forty years of attitude research had produced weak and inconsistent predictions. The 1975 move (measure attitudes at the behavior level, not the object level) was the methodological fix that made the field productive. Almost every successful applied attitude-behavior study since then is downstream of this correction.

Second: the expectancy-value decomposition. Treating attitude as a sum of belief-weighted evaluations was not original to TRA (it traces to Edwards 1954 and back through expected-utility theory), but its specific application to behavior prediction was. The decomposition is what gives the model design utility today. Without it, you have a path diagram. With it, you have a diagnostic surface.

Third: the Principle of Compatibility (TACT). The Ajzen and Fishbein 1977 paper formalizing Action, Target, Context, Time specificity is the methodological contribution most likely to outlive every other piece of TRA. Compatibility violations explain a substantial fraction of failed campaigns in behavior change, marketing, and public health. The principle is so robust that researchers who do not know they are applying it (any UX team that learns to write specific CTAs over general value statements) are still applying it.

The model is wrong in places (volitional-only scope, automatic-behavior blindness, belief-homogeneity assumption). The methodology those failures were embedded in is right. Designers who internalize the methodology and forget the path diagram come out ahead.

Where TRA Falls Apart

Every model has a working range, and the most important thing a designer can know about a model is where that range ends. TRA fails in three structurally distinct ways, each of which is worth recognizing because each implies a different fix.

1. Non-Volitional Behaviors

This is the failure mode TPB exists to fix. TRA assumes that once intention is formed, the behavior follows automatically; that the only thing standing between “I intend to quit smoking” and “I have quit smoking” is the formation of intention. This is empirically false for any behavior that requires sustained capability, resource access, social permission, biological compliance, or environmental affordance. Sheeran’s 2002 meta-meta-analysis estimated that only about 53% of intenders actually act on their intentions, with the gap widening for harder, more delayed, and less skill-mature behaviors.

For a designer, the trap looks like this. A user signs up for a financial app and indicates strong intention to save 15% of every paycheck. TRA predicts they will save 15% of every paycheck. They save 15% the first month, 8% the second, 2% the third, and zero by month four. TRA says nothing about month four because nothing in the model represents the gradient between intention and execution. TPB, RAA, COM-B, and the Health Action Process Approach all exist to fill that gap. Use TRA when the behavior is one-shot and volitional; reach for the gap-aware successors when it is recurring and capability-dependent.

2. Automatic and Habit-Driven Behavior

TRA is a model of reasoned action. The word “reasoned” is doing serious work there. The model assumes that the person is considering beliefs, evaluating outcomes, weighing referents, and forming an intention through deliberation. About half of all daily human behavior is not produced this way. It is produced by habit, which Wood and Neal (2007) defined as a learned cue-response sequence that bypasses the deliberative system entirely.

When you stop at a Starbucks drive-through on autopilot during your commute, you did not form an intention to do it that morning. The cue (specific exit on the highway) triggered the response (turn into the drive-through) directly, without a belief audit. Asking that person about their attitudes toward Starbucks coffee, their subjective norm about coffee consumption, and their perceived behavioral control over coffee budget, predicts almost nothing about whether they will stop at Starbucks tomorrow. Habit strength does.

The design move when you are working with habit-driven behavior is not to add more beliefs. It is to interrupt the cue-response sequence, redirect the cue to a new response, or change the environment so the cue does not fire. Fogg’s Tiny Habits, Wood and Neal’s context-disruption interventions, and Duhigg’s habit-loop framing all do this work. TRA does not.

3. Belief Homogeneity Assumption

The expectancy-value sum assumes that the relevant beliefs are roughly the same across the population whose behavior you are predicting, or at least that a representative belief inventory can be elicited at the population level. This works passably when the behavior is widely shared and the belief set is small (whether to vote, whether to wear a seatbelt). It breaks for behaviors where different sub-populations hold structurally different beliefs.

Consider vaccine hesitancy. A 2021 study mapped TRA across vaccine-hesitant populations and found at least three structurally different belief sets: the “rapid development” cluster (b: “the vaccine was rushed,” e: “rushed pharmaceuticals harm people”), the “natural immunity” cluster (b: “my body’s own response is sufficient,” e: “natural immunity is more durable”), and the “establishment distrust” cluster (b: “regulators have been wrong before,” e: “I should not trust an institutional consensus”). Each cluster requires a structurally different intervention. A campaign aimed at the first cluster does nothing for the third; in fact, “experts agree” messaging actively backfires for the establishment-distrust cluster.

The methodological move is to do the belief-elicitation work before designing the campaign. You cannot assume the belief inventory is shared. Fishbein and Ajzen’s 2010 book is explicit about this and walks researchers through the elicitation phase. Most applied campaigns skip the elicitation, design from a single assumed belief set, and lose the populations whose beliefs do not match.

What’s Really Happening Inside the Brain

TRA was developed before functional neuroimaging existed, so the model itself makes no claims about brain mechanism. The neuroscience that has accumulated since the 1990s does support the model’s main structural prediction, with one important refinement.

The deliberative path TRA describes (beliefs feed evaluations, evaluations sum into attitude, attitude weights against norms, intention forms, intention prompts behavior) maps reasonably well to the dorsolateral prefrontal cortex (dlPFC) and the medial prefrontal cortex (mPFC) doing executive deliberation, with the ventromedial prefrontal cortex (vmPFC) handling outcome valuation in the e_i sense. The literature on intertemporal choice, value-based decision making, and intention-to-act translation has converged on a roughly TRA-compatible neural architecture for behaviors that are genuinely deliberated.

The refinement, and this is where Kahneman’s dual-system framing matters, is that a substantial fraction of human behavior is produced by the System 1 / habit pathway, which bypasses the deliberative architecture TRA describes entirely. Wood and Neal’s 2007 neural and behavioral evidence places habitual behavior in the dorsolateral striatum, operating through stimulus-response associations that do not require the prefrontal deliberation TRA assumes. The implication is that the model’s predictive failures on habitual behavior are not just statistical; they are neuroanatomical. TRA is a model of the deliberative brain, and the deliberative brain is only sometimes in charge.

For a designer, the practical takeaway is to know which brain you are designing for. If the behavior is one the user is reasoning about (signing up for a product, choosing between options, deciding whether to adopt a tool), TRA applies. If the behavior is automatic (a habit cue firing on a scheduled trigger), you are operating in the striatal-habit pathway and the TRA toolkit is the wrong toolkit; you need habit-loop design, Charles Duhigg’s cue-routine-reward, or Wood and Neal’s context-disruption work. Different brain region, different design move.

TRA vs Other Theories

TRA’s place in the behavior-change ecosystem is best understood relative to its neighbors. The model is one node in a network of overlapping frameworks, and the differences across the network are not arbitrary; each successor was a deliberate response to a specific gap in TRA.

TRA vs Theory of Planned Behavior

TPB is TRA plus Perceived Behavioral Control. TPB therefore predicts behaviors that TRA cannot (non-volitional behaviors) at the cost of two additional survey items and a third intermediate construct. For volitional behaviors, both models predict about equally well; for non-volitional behaviors, TPB outperforms by 7-10 percentage points of variance explained. If you do not know whether your behavior is volitional, use TPB and let the data show you whether PBC is doing predictive work.

TRA vs Health Belief Model

The Health Belief Model (HBM, Rosenstock 1974, see our Health Belief Model pillar) emerged from a public-health tradition and uses different construct names (Perceived Susceptibility, Severity, Benefits, Barriers, Cues to Action, Self-Efficacy added 1988) for what is recognizably the same belief layer TRA isolates. HBM is structurally similar but has weaker mathematical formalization than TRA. HBM lacks the explicit normative-belief component, which is why HBM does poorly at predicting behaviors with strong social-pressure components (vaccination, condom use, voting) where TRA does well.

TRA vs Fogg Behavior Model

BJ Fogg’s Behavior Model (B = MAT: Behavior = Motivation × Ability × Trigger) is the design-oriented descendant of TPB. Where TPB asks “what is the belief structure under intention,” Fogg asks “what configuration of motivation, ability, and trigger produces the behavior in this moment.” The two are compatible. TRA tells you what beliefs to elicit and where to intervene. Fogg tells you how to architect the moment of action. Use TRA when you are deciding which beliefs to address in a long-form communication. Use Fogg when you are deciding what to put on a button.

TRA vs the Octalysis Framework

The Octalysis Framework is structurally orthogonal to TRA. Where TRA is a model of how beliefs combine into intention for a single behavior, Octalysis is a model of which motivational drivers a designed system is recruiting across time. The two integrate cleanly: each TRA belief is a target for Octalysis-driven design moves. A behavioral belief about “what will I get from this behavior” is the natural home for Core Drive 2 (Development & Accomplishment) and Core Drive 4 (Ownership & Possession). A normative belief about peer endorsement is the natural home for Core Drive 5 (Social Influence & Relatedness). TRA tells you which belief to move; Octalysis tells you which Core Drive to recruit to move it. This pairing is the unique-to-Yu-kai design contribution and is developed in detail in the Apply section below.

TRA in the Real World

The clearest way to see TRA’s design utility is to look at where it has been operationalized as a campaign-design instrument rather than as a survey instrument. The model has been applied across marketing, public health, electoral behavior, and product UX.

In Brand and Product Marketing

Sheppard, Hartwick and Warshaw’s 1988 meta-analysis of 87 TRA applications in Journal of Consumer Research reported an average intention-behavior correlation of r = 0.66 across brand purchase, choice, and consumption behaviors. This is unusually strong for behavioral science. The reason TRA performs so well in consumer behavior is that brand and product choices are largely volitional (the consumer can simply choose another product) and the relevant beliefs are relatively small in number and well-elicited through standard market research. A marketer asking “what does the customer believe this brand will do for them” and “which referents (peers, experts, celebrities) does the customer treat as authorities” is essentially running a TRA belief elicitation.

The applied move is to map the campaign’s job-to-be-done onto the belief layer. If the brand is underperforming because customers do not believe the product produces a specific outcome (b_i is low), the campaign should make that outcome credible. If the brand is underperforming because customers do not value the outcome the product produces (e_i is low), the campaign should reframe what makes that outcome worth caring about. Different beliefs require different creative work. The MECLABS conversion equation (C = 4m + 3v + 2(i – f) – 2a, where m is motivation and v is value clarity) is structurally compatible with this view; m and v both decompose into the same belief-weighted-evaluation sums TRA isolates.

In Public Health Behavior Change

TRA and its successors have been the dominant framework in public-health behavior change since the late 1970s. The model has been applied to condom use (Albarracin, Johnson, Fishbein and Muellerleile 2001 meta of 96 studies, average intention-behavior r = 0.45), seatbelt use, organ donation registration, blood donation, dental flossing, breastfeeding initiation, mammography screening, and Phase III vaccine adoption. The public-health domain is where the Compatibility Principle was originally stress-tested, because applied campaigns kept running null results until researchers learned to align attitude-measurement specificity with behavior-measurement specificity.

A specific cautionary example: in the 1990s, multiple national campaigns aimed at “improving attitudes toward safe sex” produced almost no behavior change despite measurable attitude improvement. The reason was compatibility violation. The campaigns measured (and improved) general attitudes toward safe sex while measuring specific behaviors (condom use in last 30 days with a specific partner). The attitudes that actually predicted the behaviors were specific to those partner-context pairs, and the campaigns never touched them. Compatibility-respecting campaigns shipped in the 2000s and produced measurable behavior change at the partner-context level.

In Voting and Civic Behavior

TRA performs exceptionally well at predicting voting behavior because voting is among the most volitional behaviors in the social repertoire (the costs of voting are low, the act takes minutes, the decision is reversible only in the next election). Studies during the 2008 and 2012 US presidential elections found that TRA-shape models predicted actual turnout with accuracy approaching 80%, where the strongest predictor was usually the subjective norm component (whether the voter’s family, friends, and identified community expected them to vote) rather than the attitude component (their evaluation of the candidates). This counterintuitive finding has shaped get-out-the-vote (GOTV) practice: peer-pressure messaging (“a record of this vote will be sent to your neighbors”) outperforms candidate-quality messaging at the margin.

In Product UX and Conversion

Conversion-rate-optimization (CRO) practice has absorbed TRA without naming it. The standard CRO move when an experience underperforms is to audit each “belief stage” the user passes through: do they believe the product solves their problem (behavioral belief), do they believe people like them use it (descriptive norm), do they believe trusted authorities endorse it (injunctive norm), do they believe they can use it without friction (capacity belief, the RAA extension). Each of these is a TRA construct. The MECLABS conversion equation, Eisenberg’s Persuasion Architecture, and the Nielsen Norman heuristic catalog are all variations on the same belief inventory.

The specific design implication is that when a conversion funnel breaks, the first investigation should be belief diagnosis, not friction reduction. Friction reduction is the right move when the failing layer is capacity belief. Outcome credibility is the right move when the failing layer is behavioral belief. Social proof is the right move when the failing layer is descriptive norm. These are not interchangeable interventions; matching the intervention to the failing belief is the actual skill.

The Elephant in the Room

Most applied behavior-change work in the past two decades has cited TRA, TPB, or RAA in the literature review and then ignored the model when designing the intervention. This is not a slight against the practitioners; it is a structural problem with how the model is taught. The model is presented as a path diagram (beliefs feed constructs, constructs feed intention, intention feeds behavior), and the path diagram is what students remember. The expectancy-value sums underneath each construct, which are the actual design surface, are presented as a methodological footnote.

What this produces in practice is interventions that target the constructs directly. A campaign to “improve attitudes toward exercise.” A program to “strengthen the norm that healthcare workers wash their hands.” A redesign aimed at “making the user feel more in control of their finances.” Each of these picks the construct as the design target and then improvises an intervention. The improvised intervention is often something general (an inspirational message, a celebrity endorsement, a feature called “Dashboard 2.0”), and the result is general: a measurable improvement on a survey item about the construct, and no measurable change in the target behavior.

The model would have predicted this. The summary construct is downstream of the beliefs; moving the construct without moving any specific belief that feeds it produces a measurement artifact, not a causal change. The way to know whether your campaign actually shifted attitude is to look at the belief inventory before and after and ask whether any individual b_i moved. If no belief moved and yet the summary attitude score went up, what you measured is response shift (the participant interpreted the question differently after the campaign) rather than attitude change.

The forty-year version of this elephant is the steady erosion of trust in attitude-survey methodology across the social sciences. The replication crisis hit attitude-behavior consistency studies hard, in part because campaigns were optimizing for the wrong layer. Researchers who stayed close to the original TRA methodology (specific behavioral beliefs, evaluated against specific outcomes, with TACT-compatible behavior measures) continued to produce replicable findings. Researchers who treated attitude as a vibe-by-survey did not. The model is not the problem. The application is.

The second part of the elephant is a designer-facing problem: behavioral product teams know about social proof, urgency, and trust signals as tactical interventions but rarely understand the belief layer they are operating on. A “social proof” component is a descriptive normative belief. A “trust signal” is some combination of expert endorsement (injunctive norm) and outcome-credibility evidence (behavioral belief). A “scarcity badge” is a behavioral belief about availability paired with a Core Drive 6 (Scarcity & Impatience) Octalysis lever. Naming the layer lets you stop using tactics by recipe and start choosing them by diagnosis.

How to Apply TRA with the Octalysis Framework

The Octalysis Framework, which I built starting in 2003 and formalized in 2012, is structured as eight Core Drives that humans respond to: Epic Meaning & Calling, Development & Accomplishment, Empowerment of Creativity & Feedback, Ownership & Possession, Social Influence & Relatedness, Scarcity & Impatience, Unpredictability & Curiosity, and Loss & Avoidance. The pairing with TRA is per-belief, not per-construct, which is the unique contribution I want to surface here. Most attempts to combine TRA with motivational frameworks end up at the construct layer (attitude, norm, PBC each get a generic Core Drive overlay), which loses the design precision the belief decomposition was supposed to give you. The leverage is one level down.

Behavioral Beliefs (b_i × e_i) → Core Drive 2 (Development & Accomplishment) + Core Drive 4 (Ownership & Possession)

Every behavioral belief has the shape “doing this behavior will produce outcome i.” That is a claim about a future state, and a designer’s job is to make the claim credible and to make the outcome feel like an asset the user is on track to own. Core Drive 2 (Development & Accomplishment) handles the credibility-of-progress side: a visible, granular representation of forward motion (a progress bar, a milestone, a streak) that converts the abstract outcome into a felt accomplishment. Core Drive 4 (Ownership & Possession) handles the asset side: framing the outcome as something the user is collecting, building up, or earning rights to. “Your savings goal” outperforms “the savings target” because the possessive pronoun makes the outcome feel pre-owned.

The design move at the belief layer is to ask, for each behavioral belief: is the outcome credible (Core Drive 2 territory), is it ownable (Core Drive 4 territory), or both. A user who does not believe the outcome is achievable needs Core Drive 2 reinforcement (proof points, progress visibility, smaller granular wins). A user who believes it is achievable but does not feel ownership of the outcome needs Core Drive 4 reinforcement (personalization, “your” framing, equity build-up). Different failure modes need different design moves, even though both are nominally “improving the behavioral belief.”

Injunctive Normative Beliefs (Authority Endorsement) → Core Drive 5 (Social Influence & Relatedness), Injunctive Mode

Injunctive normative beliefs are about what specific authority figures expect. “My doctor expects me to take the medication.” “My employer expects me to complete the training.” “My parent expects me to call on Sundays.” The design move here is to make the relevant authority’s expectation visible at the moment of decision. Note the specificity: not “people in general think this matters” (descriptive) but “this person whose evaluation you care about specifically expects this of you.”

The implementation is usually a recommendation surface featuring a relevant authority figure. Doctor recommendations on health-app onboarding. Manager-derived OKR (Objectives and Key Results) framings in performance software. Coach-prescribed goals in fitness apps. The Core Drive 5 injunctive lever fires hardest when the authority is one the user has chosen to defer to (a doctor they selected, a coach they hired) versus one imposed externally (a regulator, an institution).

Descriptive Normative Beliefs (Peer Behavior) → Core Drive 5, Descriptive Mode

Descriptive normative beliefs are about what referent peers are actually doing. “People in my building are recycling.” “Engineers at my level take parental leave.” “Households like mine donate at the giving-day push.” The design move is to make peer behavior visible at the right level of similarity. The key parameter is reference-class match: a peer is only persuasive if the user identifies them as similar enough to be informative.

This is why “10,000 people signed up” is weak descriptive evidence; it tells the user nothing about whether those 10,000 are like them. “23 people in your building signed up this week” is strong descriptive evidence; the reference class is tight. Practical applications: location-scoped social proof on landing pages, cohort-matched testimonials in onboarding, peer-aggregate dashboards in health apps. The Core Drive 5 descriptive lever fails when the reference class is too broad and succeeds when it is tight enough for the user to map themselves into it.

Control Beliefs (RAA Extension) → Core Drive 3 (Empowerment of Creativity & Feedback) for Capacity, Sludge Removal for Autonomy

The RAA split between Capacity and Autonomy maps to different design surfaces. Capacity beliefs are about “can I do this if I try” and respond well to Core Drive 3 (Empowerment of Creativity & Feedback), specifically the Bandura mastery-experience mechanism: short-loop low-stakes practice with rapid feedback. Onboarding flows that ask the user to complete a tiny version of the target behavior under tutorial conditions are building capacity belief. The 90-second guided meditation, the one-line code snippet, the single rep of the new exercise. Each is a Core Drive 3 mastery cue at the capacity-belief layer.

Autonomy beliefs are about whether the system actually lets the user do the behavior, and the design move here is mostly subtractive. Every step that exists for institutional convenience rather than user benefit is an autonomy-belief deficit waiting to happen. Sunstein’s Sludge framing (see our Sludge pillar) is the operational language for this work: walk the desired behavior end-to-end, count the friction points, remove what does not earn its place. The autonomy-belief move is not “add a Core Drive”; it is “remove an anti-Core-Drive.”

Closing the Intention-Behavior Gap → Core Drive 7 (Unpredictability & Curiosity) + Core Drive 2 + Core Drive 4 Triad

TRA stops at intention. The 47% of intenders who do not act are not failing because their beliefs were weak; they are failing because the action layer is its own design surface. The triad that closes this gap reuses three Core Drives at the moment-of-action layer rather than the belief layer. Core Drive 2 (Development & Accomplishment) shows up as visible progress on the action ladder: streaks, badges, weekly summary. Core Drive 7 (Unpredictability & Curiosity) shows up as anomaly-based cues: “your sleep dropped 22 minutes last night, unusual for you” recruits attention without asking for it, which is the if-half of Gollwitzer’s implementation intentions delivered as a notification. Core Drive 4 (Ownership & Possession) shows up as resumable partial state: half-completed forms that feel like assets you do not want to abandon.

The TRA × Octalysis Audit (Six Steps)

The audit converts the model from a vocabulary into a planning instrument. The six steps:

1. Specify the behavior at TACT resolution. Action, Target, Context, Time. Generic behaviors give you generic beliefs which give you generic interventions which give you generic null results. Get the specificity right or the rest is decoration.
2. Elicit the belief inventory from the actual population. Do not assume. Run an open-ended belief elicitation with 20-40 members of the target population, code the resulting beliefs, retain the modal beliefs as the belief inventory. Five behavioral, three injunctive, three descriptive, two capacity, two autonomy is a workable starting set.
3. Score each belief on strength and evaluation/motivation-to-comply. A 7-point scale per belief, both b and e. Compute the construct totals. Now you have a numeric baseline.
4. Identify the lowest-product belief in each construct. That is your design target. Not the construct; the specific belief whose b_i × e_i is dragging the construct down most.
5. Pick the Octalysis Core Drive that fits that belief layer. Behavioral belief → Core Drive 2 + Core Drive 4 design move. Injunctive → Core Drive 5 injunctive. Descriptive → Core Drive 5 descriptive. Capacity → Core Drive 3 mastery loop. Autonomy → subtractive design.
6. Pilot the intervention at the smallest scale that lets you re-measure the belief. Re-elicit the belief after the intervention; only ship to scale if the targeted belief actually moved.

This is the version of the model I teach inside the Octalysis Group and inside Octalysis Prime. The discipline is in the diagnosis, not in the toolkit. Most teams have the same toolkit; only some teams know which tool fits the specific belief that is failing.

Practical Steps for Applying Theory of Reasoned Action

The applied version of the model is a six-step sequence that any product team, public-health unit, or campaign group can run on any reasonably volitional behavior. The steps are deliberately small enough to fit inside a single sprint or a two-week campaign cycle.

1. Specify the behavior at TACT resolution. Write the Action, Target, Context, and Time elements explicitly. “Increase recycling” is not a behavior; “place an aluminum can in the blue bin in the kitchen on Tuesday evening” is. The TACT-resolved behavior is the only kind the model can predict.
2. Run a belief-elicitation pass with 20-40 members of the target population. Open-ended prompts: “What good things might happen if you did this behavior?” (behavioral beliefs / b_i). “What bad things?” (negative-valenced b_i). “Who would approve?” (injunctive). “Who is already doing this?” (descriptive). “Could you do this if you tried?” (capacity). “Is anyone or anything stopping you?” (autonomy). Code the answers. Retain modal beliefs as your inventory.
3. Score every belief in the inventory. Strength on 1-7. Evaluation (e_i) on negative-3 to positive-3. Motivation-to-comply (mc_i) on 1-7. Compute the products and the construct totals. Now you have a numeric baseline you can compare against after the intervention.
4. Diagnose: which belief has the lowest product (or the highest absolute negative product)? That is your target. Treat the construct total as a secondary check; the leverage is in the belief, not the construct.
5. Pick the intervention class that fits the belief layer. Behavioral belief failures → outcome-credibility evidence (case studies, mechanism explanation) and ownership framing (personalized, “your”), Core Drive 2 + Core Drive 4. Injunctive norm failure → authority-figure endorsement at moment of decision, Core Drive 5 injunctive. Descriptive norm failure → tight reference-class peer evidence, Core Drive 5 descriptive. Capacity failure → low-stakes mastery practice loop, Core Drive 3. Autonomy failure → sludge audit and friction removal.
6. Pilot, re-measure the belief, and only scale if the belief moved. Run the intervention at the smallest scale that gives you a re-measurement opportunity. Re-elicit the targeted belief in the same population. If b_i moved, you have causal evidence; ship to scale. If not, you targeted the wrong layer; return to step 4.

Six steps. Three of them (1, 2, 3) are diagnosis. Two of them (4, 5) are design. One of them (6) is causal validation. The split is deliberate. Most failed behavior-change campaigns spend their entire budget on step 5 (the intervention) and skip 1-4 entirely, which produces interventions that are tactically polished but aimed at the wrong layer. The discipline is in the diagnosis.

TRA Was the Beginning, Not the End

The Theory of Reasoned Action shipped in 1975 with a stated scope (volitional behavior) and a small set of constructs (attitude, subjective norm, intention, behavior). The fifty years since have added Perceived Behavioral Control (TPB, 1985), descriptive-vs-injunctive norm split (RAA, 2010), implementation intentions (Gollwitzer, 1999), Self-Efficacy (Bandura, 1977), habit strength (Wood & Neal, 2007), the Health Action Process Approach (Schwarzer, 2008), the Behaviour Change Wheel and COM-B (Michie, Van Stralen & West, 2011), and Sludge as the operational language of friction (Sunstein, 2022).

What survives across all of those is the belief layer Fishbein and Ajzen isolated. Every successor model can be read as adding a new construct (PBC) or a new partition (Injunctive vs Descriptive) or a new bridge to action (implementation intentions) or a new layer underneath (habit), but none of them invalidate the b_i × e_i design surface. The design surface is the durable contribution.

For a behavioral designer working in 2026, the right way to use TRA is not as a complete model but as a diagnostic instrument. You elicit the belief inventory. You score the beliefs. You identify the failing one. You pick the design move that fits that belief layer (Octalysis Core Drive at the relevant construct location). You re-measure the belief after intervention. The successor models tell you what additional layers might be in play (control, habit, action-gap closure), but the belief inventory is where the design work starts. Skip that step and you ship interventions that target the wrong layer, and the model would have predicted the null result before you ran the campaign.

Frequently Asked Questions About the Theory of Reasoned Action

What is the Theory of Reasoned Action in simple terms?

The Theory of Reasoned Action (TRA) is a model published by Martin Fishbein and Icek Ajzen in 1975 that predicts a person’s behavior from their intention to do it, where intention is predicted from two factors: the person’s attitude toward the behavior, and the person’s perception of what relevant others expect them to do (the subjective norm). What makes TRA distinct from earlier attitude theories is that both attitude and subjective norm are computed from underlying belief inventories, not entered as single scores. Attitude is the sum of (belief that the behavior produces an outcome times the evaluation of that outcome) across every outcome the person considers. Subjective norm is the same shape: sum of (belief about what a specific referent expects times the motivation to comply with that referent). The model applies to behaviors under volitional control; it was extended into the Theory of Planned Behavior (TPB) in 1985 to cover behaviors that also require capability and resources.

What is the difference between TRA and TPB?

The Theory of Planned Behavior is the Theory of Reasoned Action plus one additional construct: Perceived Behavioral Control (PBC). TRA assumes intention is sufficient to produce behavior; TPB adds PBC to handle behaviors that also require capability, resources, or environmental affordance. For volitional behaviors (voting, brand choice, signing up for a service), TRA and TPB predict about equally well. For non-volitional behaviors (sustained habits, weight loss, chronic-disease management), TPB outperforms TRA by 7-10 percentage points of variance explained. The Reasoned Action Approach (RAA), published by Fishbein and Ajzen in 2010, is the integrated successor that splits subjective norm into Injunctive and Descriptive components, splits PBC into Capacity and Autonomy components, and is what most current researchers actually use.

What is the expectancy-value formulation of attitude?

The expectancy-value formulation says that attitude toward a behavior equals the sum of belief-weighted evaluations: Attitude = Sigma (b_i times e_i), where b_i is the strength of the belief that performing the behavior will produce outcome i, and e_i is the evaluation of outcome i. The formulation traces to Edwards 1954 and the expected-utility tradition in decision theory, but its application to behavior prediction is TRA’s contribution. The design implication is that attitudes are compositional; they can be decomposed into specific beliefs that can be inspected and intervened on one at a time, instead of being treated as a single summary number.

What is the Compatibility Principle (TACT)?

The Compatibility Principle, formalized by Ajzen and Fishbein in 1977, states that attitude measures and behavior measures must match on four elements: Action, Target, Context, and Time. The acronym is TACT. If you measure attitude at one level of specificity (“attitude toward the environment”) and behavior at another (“aluminum-can recycling this Tuesday”), the model will predict poorly. Compatibility violations explain a substantial fraction of failed behavior-change campaigns. The methodological rule is to align measurement specificity: both general, or both specific, but not mixed.

What is a subjective norm in TRA?

Subjective norm is the second of the two factors TRA uses to predict intention. It captures the person’s perception of what specific referents (parents, peers, doctors, employers, identified communities) expect them to do regarding the behavior. The original TRA construct combined what later researchers separated into Injunctive Norms (what authority-bearing referents think you should do) and Descriptive Norms (what referents are actually doing). The Reasoned Action Approach split made this distinction explicit because the two operate differently: injunctive dominates when the referent has evaluative power, descriptive dominates when the referent is a peer the person identifies with.

Why does TRA stop at intention?

TRA was scoped to volitional behaviors, where the authors assumed intention was a sufficient predictor of behavior. Sheeran’s 2002 meta-meta-analysis later showed that only about 53% of intenders actually perform the behavior, with the gap growing for harder, more delayed, and more capability-dependent behaviors. This is the intention-behavior gap. The successor frameworks (TPB, RAA, Health Action Process Approach, COM-B, implementation intentions) exist to address layers of action that TRA’s volitional-only scope did not cover. For volitional behaviors the gap is small; for non-volitional behaviors it is the dominant determinant of whether the behavior actually happens.

How is TRA used in marketing?

TRA is one of the dominant academic frameworks behind brand and product-choice research. Sheppard, Hartwick and Warshaw’s 1988 meta-analysis of 87 TRA applications in consumer behavior found an average intention-behavior correlation of r = 0.66, which is strong by behavioral-science standards. The applied move is to map the marketing job-to-be-done onto the belief layer: if customers do not believe the product produces a specific outcome they value, raise the credibility of that belief; if they do not value the outcome, reframe what makes the outcome worth caring about; if they do not believe peer or expert referents endorse it, supply that endorsement at moment of decision.

What are the main criticisms of TRA?

Three structural critiques. First, the volitional-only scope excludes most behaviors of practical interest (TPB exists to address this). Second, the deliberative-reasoning assumption means TRA does not predict habitual or automatic behavior, which is roughly half of daily human activity (Wood and Neal 2007 habit research addresses this). Third, the belief-homogeneity assumption breaks for behaviors where sub-populations hold structurally different beliefs (vaccine hesitancy is the standard counterexample). Each critique points to a successor framework that handles the gap. The expectancy-value decomposition and the Compatibility Principle survive every critique and remain the durable contributions.

How does TRA combine with the Octalysis Framework?

The Octalysis Framework’s eight Core Drives map onto the TRA belief layer rather than the construct layer, which is the unique-to-Yu-kai integration developed in detail in the Apply section above. Behavioral beliefs are the natural home for Core Drive 2 (Development & Accomplishment) plus Core Drive 4 (Ownership & Possession) design moves. Injunctive normative beliefs map to Core Drive 5 (Social Influence & Relatedness), Injunctive mode. Descriptive normative beliefs map to Core Drive 5, Descriptive mode. Capacity beliefs (the RAA Self-Efficacy half of PBC) map to Core Drive 3 (Empowerment of Creativity & Feedback) via mastery-experience design. Autonomy beliefs map to subtractive design (sludge removal). TRA tells you which belief to move; Octalysis tells you which Core Drive to recruit to move it.

Is TRA still relevant today?

Yes, with the caveat that it should be used for what it was scoped for. For volitional, deliberation-based behaviors (brand choice, voting, signing up for a service, donating, choosing among options), TRA-shape models continue to produce strong predictions. For sustained or capability-dependent behaviors, the appropriate model is TPB, the Reasoned Action Approach, or HAPA. The belief-decomposition methodology and the Compatibility Principle are not optional; any behavior-change researcher or designer benefits from those regardless of which path-diagram successor they are nominally using.

References

1. Fishbein, M., & Ajzen, I. (1975). Belief, Attitude, Intention and Behavior: An Introduction to Theory and Research. Addison-Wesley. (Canonical TRA formulation.)
2. Fishbein, M., & Ajzen, I. (2010). Predicting and Changing Behavior: The Reasoned Action Approach. Psychology Press. (RAA integrated successor; Injunctive/Descriptive split, Capacity/Autonomy split.)
3. Ajzen, I., & Fishbein, M. (1977). Attitude-behavior relations: A theoretical analysis and review of empirical research. Psychological Bulletin, 84(5), 888-918. (Compatibility Principle / TACT formalization.)
4. Ajzen, I. (1985). From intentions to actions: A theory of planned behavior. In J. Kuhl & J. Beckmann (Eds.), Action Control: From Cognition to Behavior (pp. 11-39). Springer. (TPB origin paper; adds Perceived Behavioral Control.)
5. Ajzen, I. (1991). The theory of planned behavior. Organizational Behavior and Human Decision Processes, 50(2), 179-211. (Canonical TPB statement.)
6. Fishbein, M. (1967). Attitude and the prediction of behavior. In M. Fishbein (Ed.), Readings in Attitude Theory and Measurement (pp. 477-492). Wiley. (Expectancy-value attitude precursor.)
7. LaPiere, R. T. (1934). Attitudes vs. actions. Social Forces, 13(2), 230-237. (Founding attitude-behavior-gap demonstration.)
8. Edwards, W. (1954). The theory of decision making. Psychological Bulletin, 51(4), 380-417. (Expected-utility tradition that the expectancy-value formulation extends.)
9. Sheppard, B. H., Hartwick, J., & Warshaw, P. R. (1988). The Theory of Reasoned Action: A meta-analysis of past research with recommendations for modifications and future research. Journal of Consumer Research, 15(3), 325-343. (k = 87 meta; intention-behavior r = 0.66 in consumer behavior.)
10. Armitage, C. J., & Conner, M. (2001). Efficacy of the Theory of Planned Behaviour: A meta-analytic review. British Journal of Social Psychology, 40(4), 471-499. (k = 185 TPB meta; 39% intention variance, 27% behavior variance.)
11. Albarracín, D., Johnson, B. T., Fishbein, M., & Muellerleile, P. A. (2001). Theories of Reasoned Action and Planned Behavior as models of condom use: A meta-analysis. Psychological Bulletin, 127(1), 142-161. (k = 96 condom-use meta; r = 0.45 intention-behavior.)
12. McEachan, R. R. C., Conner, M., Taylor, N. J., & Lawton, R. J. (2011). Prospective prediction of health-related behaviours with the Theory of Planned Behaviour: A meta-analysis. Health Psychology Review, 5(2), 97-144. (k = 237 prospective TPB / RAA meta; 43% intention variance for RAA-shape models.)
13. Sheeran, P. (2002). Intention-behavior relations: A conceptual and empirical review. European Review of Social Psychology, 12(1), 1-36. (k = 10 meta-meta; r = 0.53 intention-behavior; the canonical “53% gap” reference.)
14. Gollwitzer, P. M. (1999). Implementation intentions: Strong effects of simple plans. American Psychologist, 54(7), 493-503. (If-then planning as the action-gap closer.)
15. Bandura, A. (1977). Self-Efficacy: Toward a unifying theory of behavioral change. Psychological Review, 84(2), 191-215. (Self-Efficacy origin; later integrated into RAA as the Capacity sub-construct.)
16. Wood, W., & Neal, D. T. (2007). A new look at habits and the habit-goal interface. Psychological Review, 114(4), 843-863. (Habit-strength model; striatal pathway; non-deliberative behavior.)
17. Michie, S., van Stralen, M. M., & West, R. (2011). The behaviour change wheel: A new method for characterising and designing behaviour change interventions. Implementation Science, 6:42. (COM-B; Capability, Opportunity, Motivation.)
18. Schwarzer, R. (2008). Modeling Health Behavior Change: How to predict and modify the adoption and maintenance of health behaviors. Applied Psychology, 57(1), 1-29. (Health Action Process Approach; Action Planning + Coping Planning gap-closers.)
19. Sunstein, C. R. (2022). Sludge: What Stops Us from Getting What We Want and What to Do About It. MIT Press. (Operational language for the autonomy-belief / friction-removal side of design.)
20. Chou, Y. (2015). Actionable Gamification: Beyond Points, Badges, and Leaderboards. Octalysis Media. (The Octalysis Framework; eight Core Drives applied per-belief in this article.)

Related Reading

• Theory of Planned Behavior: S-Tier Behavioral Designer’s Guide, the TRA successor that adds Perceived Behavioral Control for non-volitional behaviors.
• Health Belief Model: S-Tier Behavioral Designer’s Guide, the parallel public-health framework; structurally similar belief layer with weaker mathematical formalization.
• Social Ecological Model: S-Tier Behavioral Designer’s Guide, multi-level intervention map; TRA sits at the individual ring inside SEM.
• Diffusion of Innovations: S-Tier Behavioral Designer’s Guide, the population-level adoption companion; each adopter segment has distinct TRA-shape belief inventories.
• HAPA / Health Action Process Approach: S-Tier Behavioral Designer’s Guide, the intention-behavior gap closer that operates downstream of TRA.
• Sludge (Sunstein): S-Tier Behavioral Designer’s Guide, the subtractive-design lens for the autonomy-belief side.
• The Octalysis Framework, the eight Core Drives applied per-belief in this article.
• The Behavioral Framework Library, the full collection of psychological and motivational frameworks.

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