Habit Formation

What it is

Automaticity does not arrive all at once and it does not arrive on schedule. It rises steeply at first and then flattens, which means the early repetitions buy more than the later ones and the hardest stretch is the one before anything feels easier. That shape is the useful thing about the research, more useful than any particular number on the curve.

The mechanism is the transfer of initiation. Someone who trains at the same time on the same days in the same place stops deciding whether to go, because the decision was made once and Tuesday evening makes it again. Someone who trains when they feel up to it renegotiates the whole thing every session and pays the full cost of the decision each time. The second arrangement is fragile in a way that has nothing to do with how much either person wants to be fit.

The study underneath all of this is 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 to 1009. Ninety-six volunteers each chose one eating, drinking or activity behaviour to perform daily in a consistent context, and were tracked for twelve weeks with the Self-Report Habit Index. Phillippa Lally is the first author, and the finding is hers before it is anybody else's.

This is the general note under which the vault's more specialised habit material sits.

In effect

Research: the four findings, not the one

"66 days to form a habit" is a misquotation twice over. It is a median rather than a figure that applies to a given person, and the spread around it is more than eight-fold. Anyone quoting 66 days without the 18 to 254 range is quoting half a result, and the half they are dropping is the half that tells a reader whether they are behind.

Put the range to work and it changes the reading of an ordinary situation. Somebody eight weeks into a new routine who still has to decide every single time is not failing. They are inside normal, and the number that would have told them so is the one nobody prints.

The missed day, and why it is the most useful line in the paper

A single omission did not set the curve back. That is a direct empirical rebuttal to the reasoning described in all-or-nothing thinking, where a broken week reclassifies the whole effort as abandoned. The belief that one missed session ruins the thing is contradicted by the only real-world habit-formation dataset the field has, which is a stronger position to argue from than reassurance.

Everyday life: repetition is cheaper when the repetition is cheap

Habit formation and friction are the same problem seen from two ends. A behaviour that costs forty minutes of travel, a bag to assemble and a decision about what to do on arrival has to survive many more of those costs before it automatises than one that costs eleven minutes and a packed bag. Implementation intentions are the planning format that supplies the cue, and they are the standard way of handing the start of a behaviour to the environment rather than to the person. Their effect on goal attainment in general is d = .65, and for physical activity specifically it falls to somewhere around .24 to .31, which is the figure an exercise claim has to use.

What it does not say

It does not say 66 days. It says a median of 66 days with a range of 18 to 254. Those are different claims and only the second is the finding.

It does not say a habit, once formed, is durable. The paper measures how automaticity grows over twelve weeks. It does not establish that a habit survives a change of context, a house move, an injury or a new job, and nothing here should be read as saying so.

It does not generalise cleanly to exercise. The sample is small, self-selected and self-reported, and the behaviours were mostly simple. A demanding training session is at the complex end of that range, and the paper's own complexity finding says complex behaviours took longer.

It does not make identity a prerequisite. See exercise identity. In the argument this publication makes, identity arrives as a consequence of a routine that has already run, rather than as something installable before it starts.

---