The Intention-Behavior Gap: Why We Don't Do What We Say We Want

Intentions account for roughly 28% of the variance in what people actually do, which means about 72% is accounted for by something else. That figure comes from Paschal Sheeran's 2002 review in European Review of Social Psychology, a meta-analysis of ten prior meta-analyses spanning 422 studies. It is the single most useful number in behavior change research, and almost everything sold as motivation is aimed at the wrong 28%.

What the 28% actually measures

Sheeran, now at the University of North Carolina at Chapel Hill, was asking a narrow question: across the whole published literature, how well does knowing someone's intention let you predict their behavior? The answer is: better than chance, considerably worse than most people assume, and stubbornly consistent across health, consumer, academic, and social domains.

Twenty-eight percent is not nothing. Intention is among the strongest single predictors psychology has for voluntary behavior. But a variable that leaves nearly three-quarters of the variance unexplained is not a lever you can pull with confidence, and it is certainly not the thing to build a system on.

There is a harder version of this test, and it matters more. Correlational studies cannot tell you whether strengthening an intention would change anything. Thomas Webb and Sheeran ran the experimental version in a 2006 Psychological Bulletin meta-analysis of 47 experimental tests. Interventions that produced a medium-to-large shift in intention, d = 0.66, produced only a small-to-medium shift in behavior, d = 0.36. Roughly speaking, you can move what someone intends about twice as easily as you can move what they do.

Inclined abstainers: the category that explains the gap

The most clarifying thing Sheeran did was stop treating the intention-behavior relationship as a correlation and start treating it as a table. Cross intention (intend to act, do not intend to act) with behavior (acted, did not act) and you get four groups.

1. Disinclined abstainers. Did not intend to act, did not act. Intention predicted their behavior perfectly. No gap.
2. Inclined actors. Intended to act, acted. Intention predicted their behavior perfectly. No gap.
3. Inclined abstainers. Intended to act and did not. This group is where essentially the entire gap lives.
4. Disinclined actors. Did not intend to act and did anyway. Real, but rare enough to be a rounding error in most datasets.

The numbers are lopsided in a way that reframes the whole problem. Across studies of exercise, condom use, and cancer screening, the median proportion of people with positive intentions who failed to perform the behavior was 47%. The median proportion with negative intentions who acted anyway was 7%.

Read that asymmetry carefully. If someone tells you they do not intend to do something, you can predict their behavior with about 93% accuracy. If they tell you they do intend to, you are close to a coin flip. Intention functions as something near a necessary condition for action and nowhere near a sufficient one.

This also disposes of the moralized explanation. The gap is not produced by people who never cared. It is produced almost entirely by people who did care, said so sincerely, and still did not act. Whatever is going wrong is happening after the point of wanting.

Three places the gap opens

Sheeran and Webb returned to this in a 2016 review in Social and Personality Psychology Compass, which sorts the failure points rather than just measuring the shortfall. Three recur.

Getting started. The intention exists but never gets translated into a moment. There is no specified when, no specified where, and so the behavior waits for a suitably motivated version of you to show up and initiate it. Often that version does not.

Staying on track. The behavior gets started and then loses to competing demands, distractions, and better-established habits that already own the same time slot. Existing habits are the most underrated obstacle here, because they run automatically and your new intention does not.

Knowing when to stop. Less discussed and genuinely important: people persist with strategies that are not working because disengaging feels like failure. Effort continues, results do not, and the intention is eventually abandoned wholesale rather than redirected.

What the number does not mean

Two misreadings are worth heading off, because both are common in productivity writing that cites this literature.

The first is "intentions do not matter." They plainly do. They are a strong predictor by the standards of the field, they are close to necessary, and interventions that raise intention do move behavior, just less than proportionally. The correct conclusion is that intention is a poor foundation, not that it is irrelevant.

The second is that 28% is a precise constant. It is an average across heterogeneous studies with heterogeneous measures, many of which relied on self-reported behavior. Self-report inflates the apparent correlation, because people's accounts of what they did are contaminated by what they meant to do. Studies using objective measures generally find weaker relationships. If anything, 28% is the optimistic figure.

It is also worth naming what does not explain the gap. The familiar story is that willpower is a limited resource that runs out, so evening intentions collapse. That model, ego depletion, failed a 23-laboratory preregistered replication in 2016 involving 2,141 participants, which found essentially no effect. Depletion is not a load-bearing explanation for anything here, and much of the habit writing that still invokes it has not kept up with the evidence.

What actually narrows it

The interesting consequence of the inclined-abstainer finding is that the intervention target is obvious. You do not need to make people want it more. You need to convert an existing want into a specified occasion.

That is precisely what if-then implementation intentions do, and it is why they are the most tested response to this problem in the literature. Adding an obstacle-identification step produces Gabriele Oettingen's four-step protocol, laid out in the WOOP walkthrough. Neither is dramatic. Both address the actual failure point rather than the one that is easier to sell.

The broader implication is the one covered in the comparison of motivation against affirmation as behavior change strategies: a system that requires you to want it enough on the day will fail on roughly half the days you wanted it. Design for the inclined abstainer, because on any given morning that is who you are.