Habit formation

Why does a behaviour get easier to repeat and harder to stop the more often it happens?

Ask someone why they checked their phone and you will often get a pause, then a shrug. They did not decide to. Something more interesting than forgetfulness is going on there: at some point a behaviour that was chosen stopped being chosen, and started merely happening. And notice the asymmetry the question points at — repetition does not only make a behaviour easier, it makes stopping it harder, which is not the same thing. What could change with repetition that would produce both effects at once?

Start with what a new behaviour requires. You must represent the outcome you want, work out what action produces it, and judge whether it is still worth doing — attention, working memory, and a moment of evaluation each time. Call it goal-directed control: the action is performed because of the outcome it is expected to bring.

Now ask a designer's question. If a situation recurs many times and the same action reliably produces the same acceptable result, why re-run that evaluation every time? A cheaper arrangement is available: bind the situation directly to the action and skip the middle. Perceive the cue, run the routine. That is stimulus-driven control, and it costs almost nothing.

So the plausible story is that repetition shifts a behaviour from the first system to the second. Is there evidence, or is that just a tidy hypothesis? There is a clean experimental test, and it is the heart of the field. Train an animal to press a lever for a food reward, then devalue that reward — feed it to satiety, or pair it with mild illness so it is no longer wanted. An animal with modest training stops pressing almost immediately: it was pressing for the food, and the food is no longer worth having. An overtrained animal keeps pressing anyway. The behaviour has come loose from the outcome that built it. That insensitivity to outcome value is the working definition of a habit, and it is measurable rather than metaphorical.

That result explains the asymmetry we started with. Easier to repeat, because evaluation is skipped. Harder to stop, because "I no longer want the result" is an argument addressed to a system that is no longer in charge.

What is the cue? Whatever reliably preceded the routine — a time, a place, a preceding action, an emotional state. This is the cue-routine-reward loop, and the reward's role is worth stating carefully: it is what caused the association to be learned, and it need not still be delivering anything for the loop to keep running. In the brain, the shift is associated with a change in which striatal circuits control the behaviour — from regions tied to outcome value toward regions tied to stimulus and response — though the human evidence for that mapping is much thinner than the animal evidence.

Now the practical question: how long does it take? Here a specific correction is needed, because the popular answer is wrong. The "21 days to form a habit" figure traces to a plastic surgeon, Maxwell Maltz, who observed in 1960 that his patients seemed to take about three weeks to get used to their new appearance, and wrote that it took "a minimum of about 21 days" for an old mental image to dissolve. That was an anecdotal clinical observation about adjusting to a changed face, not a study of habit formation — and it lost the word "minimum" on its way into folklore. When it was actually measured, in a 2010 study by Lally and colleagues that tracked people adopting a new daily behaviour, the median time to reach a plateau of automaticity was around 66 days, with individual estimates ranging from about 18 days to well over 200. The spread is the finding. Simple behaviours in stable contexts consolidated fast; harder ones took months.

Three refinements that connect the pieces.

First, "goal-directed" and "habitual" are not two kinds of person or two kinds of behaviour, but two systems that run in parallel and compete for control of the same action. Which one wins depends on training, on how stable the context is, and on load — stress, time pressure, fatigue and distraction all tilt control toward the habitual system, which is exactly why resolutions fail on bad days rather than average ones.

Second, habits are not erased. The evidence from extinction studies is that the old association survives learning a new one, and resurfaces under stress or on return to the original context. This is why the durable strategies are contextual rather than heroic — change the cue, change the environment — and why life disruptions such as moving house are unusually good moments to change behaviour: the cues that carried the old routine are simply absent.

Third, the loop is not a theory of everything. Cue-routine-reward describes the acquisition of relatively simple repeated actions well, complex goal-pursuit much less well, and its popular versions — the 21-day figure among them — have run far ahead of what the research supports.

• Why implementation intentions — deciding in advance which cue triggers which action — outperform resolutions.
• How compulsion in addiction relates to, and differs from, ordinary habitual control.

Every term the collection defines is gathered in the glossary.