A student chooses to put a glass of water beside breakfast each morning. On Monday it takes a reminder. Weeks later, reaching for the glass may feel less like a fresh decision. How long did that transition take, and what exactly changed? The popular answer is a single number of days. Phillippa Lally and colleagues asked a narrower, better question: how does reported automaticity develop when people repeat a chosen action in a stable everyday context?
Their 2010 study is useful because it followed people outside a laboratory over twelve weeks and modeled each person’s trajectory. It is also easy to overstate. The participants chose simple eating, drinking or activity behaviors; the main outcome was a self-report measure; and the best-known time estimate came from a subset whose data fitted the chosen curve well. A research brief should keep those design details attached to every practical conclusion.
The question behind the famous number
Habit is often treated as a synonym for doing something many times. In psychology, the more precise interest is whether a recurring context begins to cue a response with less deliberation. Someone may exercise every Tuesday through considerable effort and still not experience much automaticity. Another person may reach for a familiar snack when a particular meeting ends, even after deciding to eat differently. Frequency and automaticity are related but not interchangeable.
Lally and colleagues wanted to observe the formation side of that process. Rather than asking people to recall how long a habit took, they invited volunteers to begin a new daily behavior tied to a cue such as eating lunch or arriving home. They then gathered repeated reports of whether the action occurred and how automatic it felt. That design follows change over time within the same person, but it cannot remove all the other things that change in ordinary life.
The paper’s question was not “How long does every habit take?” It was whether a plausible learning curve described the rise in automaticity for these selected behaviors and people. It also examined how consistent performance and missed opportunities related to that curve. The study did not randomize participants to different durations of practice, settings or cue types. Its estimates are best read as modeled descriptions of this field sample.
What the researchers did
Ninety-six volunteers chose an eating, drinking or activity behavior to perform daily in the same context for twelve weeks. They reported each day whether they had performed it and answered items drawn from the Self-Report Habit Index concerning the action’s automatic character. The context link mattered: repetition after the same cue was the mechanism the researchers hoped to observe, not merely a high total number of repetitions.
The researchers modeled how each person’s automaticity score changed across the 84-day observation period. An asymptotic curve rises at first and gradually approaches a plateau. In this context, that shape was a hypothesis about learning: each performance could add less to felt automaticity than the preceding performances. The curve does not imply an invisible switch that suddenly turns a deliberate action into a habit on one particular date.
Attrition and model fit narrow the inference. Eighty-two participants supplied enough data for analysis. The proposed model could be fitted for 62 individuals, and it described 39 of those particularly well. The often-cited median of 66 days to reach 95 percent of the modeled asymptote comes from that good-fit subset. The estimated range there was 18 to 254 days, with the upper estimate extending beyond the twelve weeks actually observed. This is not a clock read directly from all 96 volunteers.
That distinction changes the headline. “The median modeled time among the good-fit cases was 66 days” is faithful to the paper. “A habit forms in 66 days” removes the sampling, measurement and modeling conditions that give the number meaning. The 18-to-254-day range also does not provide a reliable personal forecast. Behavior difficulty, context, starting motivation and opportunity differed, and the small good-fit group limits precision.
What changed during repeated action
For many participants, reported automaticity rose with repeated performance and tended toward a flatter trajectory. This pattern fits a cue-response account of habit acquisition: in a stable situation, an action can become easier to initiate. Wood and Rünger’s later review places this learning in a broader psychology of habit, describing how repeated goal-directed behavior in recurring contexts can produce more default responses. The review helps interpret Lally’s result, but it is not an additional observation from the same 2010 participants.
The selected behaviors were not equivalent. Taking a drink with lunch and beginning a longer walk impose different demands. Even if both are repeated daily, their opportunities differ: travel, weather, health, work schedules and other people can interrupt one more than the other. A plausible practical inference is to make the desired action small enough and the cue clear enough to be encountered regularly. The 2010 paper does not establish one ideal action size or guarantee that every person benefits from the same cue.
The result also complicates the language of discipline. Early performance can require deliberate planning while the association with a cue develops. That does not make the action less valuable. Nor does a later feeling of ease mean a person has lost agency. A habit can serve a chosen goal by reducing the effort of starting; the person can still review whether the goal remains worthwhile and alter the context if it does not.
Wood and Neal’s account of the habit-goal interface makes this relationship explicit. Habits and reflective goals can cooperate, but familiar cues may also elicit actions after a person has changed their mind. The useful question is therefore not whether every response should become automatic. It is which repeated actions deserve that efficiency, and which situations require a fresh decision.
What a missed day means
Lally and colleagues examined missed opportunities in their data. A single missed occasion did not materially disturb the modeled formation process in the way a strict streak rule would suggest. This observation speaks to one omission in this setting, not to every interruption or a license to disregard practice. A shift change, illness or loss of access to a cue can alter the situation more deeply than missing Tuesday’s chosen action.
Streak-based apps may treat one gap as a reset because a continuous count is simple to display and may encourage return. The study’s result provides a reason to be cautious with that story. If a person misses a day, the practical task may be to find the next real opportunity, not to declare that earlier learning disappeared. This is an Institute application of the result, not a tested app intervention in Lally’s study.
At the same time, a person who repeatedly misses the chosen action may not have a stable opportunity. A reminder cannot create a lunch break, safe walking route or accessible exercise option. Repetition is a learning process situated in conditions. A program that blames individuals for missed cues without checking those conditions mistakes a structural obstacle for weak commitment.
How automaticity was measured
The researchers used self-reported items about acting automatically. These ask a person to reflect on how little thought or effort seems involved, a sensible but imperfect way to study a private experience across many days. Self-report can change with expectations, memory and a participant’s understanding of the item. It does not directly record a neural pathway or prove that the action would survive a major context change.
Gardner and colleagues later examined an automaticity subscale of the broader Self-Report Habit Index. Their analysis helps explain why researchers often focus on automaticity items rather than treating frequency and identity as the same construct. Yet validating a shorter questionnaire in several datasets does not convert any single participant’s score into an objective diagnostic test. A high score says what the person reported under a defined measure.
Other evidence can complement self-report. Repeated observation could record whether the action occurs when its cue appears. A context change could test whether the action follows the old cue or adapts to a new one. A person can also explain what they intended and how much effort was involved. These measures answer different questions, and none alone gives a complete account of motivation, skill and habit.
Orbell and Verplanken’s experimental and field work on cue-contingent automaticity illustrates that a cue can be central. Their studies linked habit measures to attention to smoking cues, action slips in a changing policy context and planned flossing in response to a specified situation. These are distinct behaviors and designs. They support a general interest in context-response links, not a claim that every everyday routine is produced by the identical mechanism.
What the paper changed in public understanding
The 2010 study made it harder to defend a universal deadline for habit formation. It offered a more realistic picture: often gradual, variable and dependent on repeated action in context. The influential 66-day figure is useful if it prompts a person to expect that automaticity may take longer than a brief challenge. It misleads if it becomes a promise that a particular action will feel effortless on day 67.
The 21-day slogan is especially poorly supported by this paper. Lally and colleagues did not find a uniform threshold at three weeks, and the good-fit estimates extended far beyond it. A person who has to make a deliberate choice after 21 days has not failed an empirical test. The appropriate comparison is with their own earlier experience in a stable context, while remembering that self-ratings can be noisy.
The study also illustrates the value of looking at a trajectory rather than one before-and-after score. If automaticity grows steeply at first and then slows, a two-point comparison could miss the shape. If some people show little growth, an average curve could hide them. Reporting who contributed to the best-fitting model guards against turning an attractive graph into an unqualified account of everyone.
It is tempting to turn the rising curve into a motivational chart for a new participant. That would use group-derived model estimates as if they were a personal progress meter. A score that flattens may reflect a genuinely easier routine, but it may also reflect a stable response style on a questionnaire. A score that fluctuates could reflect a change in timing or context. In a service, the more useful conversation may ask what the person was trying to make easier, what cue was available and whether the action remains welcome. Those details can guide a revision even when the curve alone is ambiguous.
Why this matters in Northern Europe
Nordic schools, workplaces and public health services often support long-term actions such as active travel, medication routines or regular practice. A habit lens can help them ask whether a routine is feasible in the places people actually live. Winter conditions, shift work, transit access and care responsibilities may determine whether a cue can be met reliably. These are contextual examples, not variables tested by Lally and colleagues.
A service could make an agreed action easier to repeat by choosing a cue with the person, offering an accessible alternative and allowing a plan to change when circumstances do. It could evaluate the person’s experience of automaticity and the real behavior separately. An app may record a streak while the action still feels difficult; another person may act easily without reporting every instance. Both possibilities matter when public services decide whether a tool is helping.
There is an ethical boundary. A system can use cue design to support a person’s chosen aim or to capture repeated engagement for its own metric. The fact that repetition can become easier does not justify hiding defaults, escalating reminders after refusal or treating access barriers as personal defects. The person should be able to understand the proposed routine, change it and stop it.
What the evidence does not show
Lally et al. did not demonstrate a universal time to make a habit. The median modeled estimate of 66 days applies to the 39 good-fit participants, not the original 96. The 18-to-254-day range came from model estimates, including extrapolation beyond observation for slower trajectories. The study’s self-report outcome does not establish that every participant’s behavior became effortless, permanent or independent of context.
The paper did not compare different habit apps, educational methods or national public health programs. Its single-missed-opportunity analysis does not tell us what repeated gaps, unsafe conditions or major life changes do in every case. It did not establish that a habit is desirable merely because it becomes automatic. That judgment requires a person’s goals, circumstances and possible costs.
Nor should later habit reviews be folded into the 2010 data. They explain mechanisms and measurement questions across a wider literature. The Institute’s suggested practice is to choose a meaningful action, locate a reliable cue, make repetition feasible, treat occasional gaps without moral drama and review whether the routine still serves its purpose. That is a cautious synthesis, not a protocol validated by Lally’s one study.
Return to breakfast
The glass by breakfast may become easier to reach for as the morning context repeats. The student may notice the action earlier, remember it more often, or eventually do it without an explicit reminder. There is no need to predict a particular day. If breakfast changes, the cue may need to change too. The study invites a patient question: what happens to this action, for this person, in this setting over time?
A habit is not a calendar prize awarded after a streak. It is a possible change in how context and action become linked. The most useful lesson of Lally et al. is therefore a methodological one as well as a practical one: observe the process, keep the conditions visible and let variation remain visible when the headline asks for one simple number.
Research references
- Lally, P., van Jaarsveld, C. H. M., Potts, H. W. W., & Wardle, J. (2010). How are habits formed: Modelling habit formation in the real world. European Journal of Social Psychology, 40(6), 998–1009. View source
- Wood, W., & Neal, D. T. (2007). A new look at habits and the habit-goal interface. Psychological Review, 114(4), 843–863. View source
- Wood, W., & Rünger, D. (2016). Psychology of habit. Annual Review of Psychology, 67, 289–314. View source
- Gardner, B., Abraham, C., Lally, P., & de Bruijn, G.-J. (2012). Towards parsimony in habit measurement: Testing the convergent and predictive validity of an automaticity subscale of the Self-Report Habit Index. International Journal of Behavioral Nutrition and Physical Activity, 9, 102. View source
- Orbell, S., & Verplanken, B. (2010). The automatic component of habit in health behavior: Habit as cue-contingent automaticity. Health Psychology, 29(4), 374–383. View source