Habit formation statistics: every number, and its source

Updated 26 August 2026

• The 66-day figure is a median, not a duration; the same study reported individual results from 18 days to 254, and one participant had not levelled off when the modelling stopped.
• The 21-day rule comes from a 1960 book by a plastic surgeon describing how long patients took to accept their new appearance. No study produced it.
• The best-evidenced action is not repetition or tracking. It is stating in advance when and where the behaviour will occur, at d = 0.65 across 94 tests.
• Missing one opportunity had no measurable effect on automaticity in the study that looked.
• Behaviour complexity mattered and willpower was not measured to matter: simple actions settled faster than effortful ones.
• No credible figure exists for what share of habits fail, because failure has never been defined consistently enough to count.

How long it takes: the 66-day study

One field study is the source of essentially every honest claim about duration. Ninety-six volunteers each picked a single eating, drinking or activity behaviour, performed it daily in the same context, and reported each day for twelve weeks. Eighty-two supplied enough data to analyse.

The researchers fitted an asymptotic curve to each person’s automaticity ratings across the 84 days and asked when that curve reached 95 per cent of its own plateau. The middle answer was 66 days.

The spread is the part worth carrying: 18 days at one end, 254 at the other. The upper figure marks where the modelling stopped rather than where that participant finished, and at least one curve had not flattened by the end of the study. A single number quoted from this work is not a simplification of it; it is the discarding of its main result. The full argument is in how long it takes to form a habit.

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. doi:10.1002/ejsp.674

What actually helps: d = 0.65

The largest effect anyone has measured for a thing you can decide to do is not repetition, tracking or motivation. It is naming the occasion in advance — a plan of the form “when situation X arises, I will do Y”, which the literature calls an implementation intention.

A meta-analysis pooled 94 independent tests covering more than 8,000 participants and reported a medium-to-large effect, d = 0.65, on whether the intended behaviour happened. That gain sat on top of whatever motivation people already had, which is the useful part: it is available to someone who is not feeling especially determined.

This is why choosing the moment is a larger decision than choosing the size, and why an app that asks only what you intend to do has skipped the better question.

Gollwitzer, P. M., & Sheeran, P. (2006). Implementation intentions and goal achievement: A meta-analysis of effects and processes. Advances in Experimental Social Psychology, 38, 69–119.

What breaks them: circumstances, not character

Habits are anchored to the setting they were built in. When the setting moves — a new house, a new shift pattern, a holiday — the cue that used to fire stops arriving, and the behaviour lapses without any decision having been made to stop.

That is a more useful account of a lapse than a personal one, because it points somewhere. A habit that fell over during a house move needs re-attaching to something in the new house. An irregular schedule is the same problem arriving weekly.

Wood, W., Tam, L., & Witt, M. G. (2005). Changing circumstances, disrupting habits. Journal of Personality and Social Psychology, 88(6), 918–933.

The numbers that do not exist

This is the section most statistics pages leave out, and it is the more useful half.

21 days. Traceable, and not to research. It appears in Psycho-Cybernetics, published in 1960 by Maxwell Maltz, a plastic surgeon, who noted that patients seemed to need about three weeks to stop seeing their former face in the mirror. An observation about adjusting to a changed appearance became, through repetition, a law about habits. It survives because three weeks is a marketable length for a programme and 66 days is not.

A failure rate. There is no defensible percentage of habits that fail, and the figures in circulation cite each other rather than a study. Two things would be needed and neither exists: an agreed definition of failure, and a representative sample followed long enough to observe it. A habit abandoned at week three and one performed four days in seven are not obviously the same outcome, and no dataset settles which counts.

A count of how many habits a person has. Sometimes given as a share of daily behaviour. The underlying research measures how often people report acting in stable contexts, which is a different quantity, and the conversion into a tidy percentage happens outside the papers.

Anything about habit apps specifically. Retention and success figures for this category come from the companies selling the apps. Treat them as marketing, including ours: we publish no such number, because we would be the only source for it.

The method is simple enough to apply anywhere. Ask which study, how many people, what was measured, and over what period. A statistic that cannot answer those four is a sentence that sounds like a statistic.

Why this page has so few numbers

A page called statistics is expected to be long, and the ordinary way to make it long is to include figures whose provenance nobody checks. That trade is available and we have declined it, because the value of a reference page is entirely the trust in it and one unsourced figure removes that from every other line.

So the count here is small on purpose. If a number is missing, it is because we could not follow it to a paper that measured it.