Identity-Based Habits: The Science of Who You Become

“I’m trying to quit” and “I’m not a smoker” describe the same behavior. Only one of them makes a claim about who you are.

James Clear built the most-quoted chapter of Atomic Habits on that distinction: don’t chase outcomes, become the type of person who gets them. It’s a good idea — and unlike most habit advice, parts of it rest on real randomized experiments. Here’s the honest map. Framing a behavior as an identity (“be a voter,” “don’t be a cheater”) has shifted real behavior in controlled studies, including one measured against official state voting records. One of those findings then became the center of a famous replication fight. And the bigger claim — that adopting an identity produces durable habit change over months — is plausible, supported by correlational data, and not yet proven by any long-term trial.

Where does the idea come from?

Clear popularized the term, but the mechanism is older self-concept psychology: people work to keep their behavior consistent with the identities they value. If “voter,” “runner,” or “honest person” is part of your self-image, acting against it carries a cost that “skipping a task” doesn’t. The interesting question is whether nudging that self-image actually changes what people do. That’s been tested.

Is there real evidence that identity framing changes behavior?

Yes — and it’s the part of the identity-habits story that the summary blog posts almost never cite.

A 2011 paper in PNAS (Bryan, Walton, Rogers & Dweck) ran three randomized experiments that varied a single word class in a pre-election survey: voting framed as an identity (“being a voter”) versus an action (“voting”). The noun phrasing significantly increased interest in registering to vote in experiment 1 — and in experiments 2 and 3, run around the 2008 US presidential election and the 2009 New Jersey gubernatorial election, it increased actual turnout as verified by official state records. The reported gap was large: roughly 11 to 14 percentage points higher turnout in the noun condition.

The same team then flipped the polarity. A 2013 study (Bryan, Adams & Monin) in Journal of Experimental Psychology: General gave participants a chance to claim money they weren’t entitled to. Instructions said either “please don’t cheat” or “please don’t be a cheater.” Across three experiments, the “cheating” group claimed significantly more money, while the “cheater” group showed no evidence of having cheated at all — even in a private online setting where nobody was watching. A 2020 preregistered replication (Guo et al., 768 participants) using a coin-toss paradigm in Japan replicated the effect: the identity wording (“don’t be a cheater”) curbed dishonest reporting where the action wording didn’t.

So the core mechanic is real: when a behavior is framed as evidence about who you are, and that identity matters to you, decisions move.

Didn’t the voter study fail to replicate?

Partly — and this is the boundary most write-ups skip in both directions.

In 2016, Gerber and colleagues published a large field experiment in PNAS — more than 14 times the combined sample of the original studies — and found no significant difference between the noun and verb framings; the point estimate was approximately zero. Both framings did worse than a shorter, standard get-out-the-vote message.

Bryan’s team answered in a 2019 PNAS analysis, and the substance matters more than the score-keeping. The original experiments ran the day before and the morning of high-profile elections, when “being a voter” is a live, valued identity. The replication ran across four days before mostly uncompetitive August midterm primaries — fewer than 10% were competitive races. Their argument: the effect is context-dependent, not imaginary — identity framing works when the identity is salient and worth claiming, and their reanalysis of a second, higher-salience replication found the effect did appear under those conditions.

The fair reading: this is a live scientific debate, not settled doctrine. The effect has replicated in some contexts (cheating, high-salience elections by Bryan’s analysis) and flatlined in others. Identity framing is a conditional lever, not a universal one.

Do habits actually become part of who you are?

This is the other half of Clear’s claim — that repeated behavior feeds back into identity — and here the evidence is real but correlational.

Verplanken and Sui (2019), in Frontiers in Psychology, had people rate 80 behaviors for both habit strength and whether each felt like part of their “true self.” The two tracked together, and the link was strongest when the habits served values the person explicitly cared about. Stronger habit-identity integration also went with higher self-esteem and a pull toward an ideal self. The authors’ conclusion is carefully worded: linking habits to identity may sustain newly formed behaviors — a hypothesis, not a demonstrated effect.

A 2025 twelve-week study (Alfrey, Condie & Rebar, N=98) tracked identity, intention, motivation, habit, and behavior weekly. Identity predicted behavior — but much of the within-person effect ran through intention, and identity itself barely moved across the twelve weeks. The authors’ phrase: identity “may be difficult to change.” You don’t declare a new self on Monday and have it hold by Friday.

So do identity-based habits work?

Split the claim in two and grade each honestly.

Proven: identity language changes specific, discrete decisions in randomized experiments — registering, turning out, not cheating — when the identity is one you value and the moment makes it salient. That’s more experimental support than most habit advice ever gets.

Not proven: the full Atomic Habits arc — decide you’re “a runner,” and months of durable behavior follow. No long-term RCT has tested identity adoption against a control for habit maintenance. What we have instead is a consistent correlational picture (habits tied to values feel like the self, and identity predicts behavior over weeks) plus a mechanism that works in single moments. Reasonable to act on. Wrong to call settled.

How to use it without fooling yourself

• Use noun language at the decision point, not as a mantra. “I don’t smoke” when offered a cigarette; “I’m someone who trains” when the alarm goes off. The experiments moved single decisions, so deploy the framing where decisions happen.
• Attach the habit to a value you already hold. That was the moderator in Verplanken and Sui’s data — habits felt like the self when they served something the person cared about. An identity you don’t value has nothing to protect.
• Let evidence build the identity, not the reverse. Identity was the slowest-moving variable in the 2025 data. Stack small completed reps as proof — the mechanics in habit stacking and the honest timelines in how long habits take to form do the accumulation work.
• Don’t let one lapse revoke the identity. A missed day is an event, not a verdict — and self-compassion after a lapse predicts getting back on track better than self-punishment does.

Identity is also where hypnosis-based behavior change does its work: rehearsing the self-image that makes the behavior feel like yours, rather than white-knuckling the behavior itself. For how that fits the broader evidence, see the pillar guide on AI hypnotherapy and behavioral change.