Self-Reported Habit Strength

What it is

The measure comes in three families, and the distinction between them decides what a given finding means. The first is behavioural frequency multiplied by context stability, from Ouellette and Wood's 1998 review and Wood, Tam and Guerrero Witt's 2005 field study. The second is the response-frequency measure developed by Bas Verplanken and colleagues in 1994. The third is the Self-Report Habit Index of Verplanken and Sheina Orbell, 2003, and its derivatives, which ask directly about automaticity with items about doing something without thinking.

A single frequency item is the thinnest version and it appears in the most-cited field study in the literature. David Neal, then at the University of Southern California, with Wood, Mengju Wu and David Kurlander, gave cinema-goers popcorn popped an hour earlier or seven days earlier, with boxes weighed on a digital scale before and after and an average starting weight of 61.73 grams. Habit strength was, in the authors' own words, one item asking participants to indicate on a seven-point scale how frequently in the past they ate popcorn in movie theatres. Strong-habit participants in the cinema ate about sixty-three per cent of the box regardless of palatability, the only absolute percentage the paper prints in its running text. The construct in the headline is automaticity. The measure in the method is recalled frequency.

The most useful evidence that the inference does not hold comes from inside the same laboratory. Neal, Wood, Labrecque and Lally's 2012 study found that participants with stronger running habits reported their running was goal-driven at r(51) = .54, and that this belief was uncorrelated with the implicit context-cue measure at r(51) = .07. Their stated conclusion is that research on the actual mechanisms of habit performance cannot rely on judgments and inferences as tests of underlying mechanisms, and that research tapping the subjective experience of habits is especially likely to find that habit performance depends on goals.

The field's own synthesis then supplies the result that makes the doubt quantitative. Hagger and colleagues, pooling 267 independent studies and 107,813 participants in 2023, compared the three measure families and report larger habit-behaviour effects in studies whose self-report habit measures included frequency items than in those that omitted them. They close by saying their findings are expected to catalyse future habit research using experimental methods and non-self-report measures. A measure that contains past frequency predicts behaviour better than one that does not, which is what you would expect if part of what the instrument captures is simply repetition predicting repetition.

The objection has also been made in print and answered in print. Falko Sniehotta and Justin Presseau published a letter in Annals of Behavioral Medicine in 2012 arguing that where a habit operates outside awareness, a self-report is an inference about one's own behaviour rather than a report of the habit, and Benjamin Gardner, Charles Abraham, Phillippa Lally and Gert-Jan de Bruijn replied in the same issue. Neither letter nor the index's own founding paper was read for this page, and all three are cited for their existence only. The critique above does not depend on them: it rests on the 2012 dissociation and the 2023 meta-analytic result, both read at source.

In effect

The popcorn box, weighed to the gram

The research case is worth setting out as a contrast. The outcome measure in the popcorn study is excellent: grams of popcorn eaten, weighed on a scale, with freshness manipulated and a liking check confirming the manipulation worked. The predictor is one remembered rating. A study can be rigorous on its dependent variable and thin on its independent one, and this is the clearest instance in the habit literature. The three-way interaction of habit strength, context and freshness came in at p = .043, and when liking was added as a covariate in the second study the critical interaction fell below significance, which the authors report and attribute to low power.

Answering "how automatic is this for you?" about brushing your teeth

The everyday version is to try answering the question the index asks. Most people can say roughly how often they brush their teeth. Far fewer can say whether they attended to it this morning, because the thing being asked about is precisely the thing that leaves no trace worth recalling. The answer people give is reconstructed from frequency, which is exactly the circularity the meta-analysis detected: the items people can answer are the ones that carry the predictive weight.

Engagement dashboards and the same error at scale

A product team measuring daily active users, session frequency and streak length is measuring repetition and often calling it habit. The same gap applies, with the advantage that the behaviour is observed rather than recalled and the disadvantage that no part of the instrument touches automaticity. The tests that would are available and rarely run: latency to begin after a cue appears, whether the behaviour survives the reward being withdrawn, or whether it persists when the cue is moved. Repetition is cheap to instrument, and automaticity is not, which is how the measurement problem reproduces itself outside the academy.

What it does not say

It does not say the measures are useless. They are predictive, widely used and pooled across more than a hundred thousand participants, and that record is real.

It does not say the measures reach automaticity. That is the specific claim the field's own meta-analysis gives reason to doubt, and the doubt comes from the direction of the finding rather than from an absence of data.

It does not say the three instrument families are interchangeable. A single frequency item, a frequency-by-stability product and a twelve-item automaticity index are different measures, and a headline rarely says which produced it.

It does not settle the published exchange about the index. The objection and the reply are both in the record and neither was read for this page.

Who developed it

The frequency-by-context-stability measure comes from Judith Ouellette and Wendy Wood's 1998 review and Wood's later field work. The Self-Report Habit Index is Bas Verplanken and Sheina Orbell's, from 2003. The dissociation that undercuts the construct claim is David Neal's first-authored 2012 paper with Wood, Jennifer Labrecque and Phillippa Lally, and the meta-analytic result is Martin Hagger's with seven co-authors.

---