The Intention-Behaviour Moderator

What it is

The review was built on sixty separate research reports yielding 64 independent studies, searched across PsycLIT and Sociofile for 1974 to 1994, Dissertation Abstracts for 1867 to 1994 and Psychological Abstracts for 1920 to 1973. Inclusion required a past-behaviour frequency measure and an attitude measure on a corresponding object. The headline correlations are modest: past behaviour to intention at r = .43 across 33 study outcomes, past behaviour to future behaviour at r = .39 across 16.

The moderator is the finding. In the authors' own words, frequency of past behaviour was a weaker predictor of future behaviour in domains encountered only annually or biannually presenting unstable contexts, at r = .27, than in domains encountered daily or weekly presenting stable contexts, at r = .59, with a z of 8.98. The gap between those two correlations is larger than almost anything else in the literature, and it is a gap between kinds of behaviour rather than between kinds of person.

The path models decide whether the popular reading, that intentions stop mattering once something is a habit, is fair. They say it is not. In low-opportunity unstable domains, six study outcomes covering blood donation, voting, flu shots and mammograms, past behaviour carried a standardised weight of 0.12 and intention 0.62. In high-opportunity stable domains, eight study outcomes covering exercise, fast food, class attendance, seat belt use, drinking milk and church attendance, past behaviour carried 0.45 and intention 0.27, with a confidence interval from 0.20 to 0.34. Intention's weight more than halved. It did not reach zero, and the authors wrote only that conscious intentions had a less marked impact on future responses.

The finding has held up, which is more than most results in this area can claim. Martin Hagger, at the University of California Merced and the University of Jyvaskyla, led a 2023 meta-analysis in Motivation Science, preregistered at PROSPERO, pooling 243 articles, 267 independent studies and 107,813 participants across diet, physical activity, transport, technology use, protective behaviours, alcohol, smoking and medication adherence. It found small to medium non-zero average effects of both habit and intention on behaviour, with habit partially mediating the effect of past behaviour and accounting for roughly a quarter of its total effect, and a large residual past-behaviour effect left over. Larger habit-behaviour effects appeared where the behaviour had high opportunity for habit formation and low complexity, which is the 1998 moderator restated at twenty-five years' distance. The numeric tables did not render in the accepted manuscript read for this page, so no coefficient from that paper is printed here.

The modern reframing points the same way from a different instrument. A 2025 ecological momentary assessment study, prompting 105 people six times a day for a week, reported about two thirds of behaviours habitually instigated, about seventy-six per cent intentional and forty-six per cent both. Those shares were never meant to total a hundred. Most behaviour is habitual and intended at the same time, largely because habits are built out of things a person once meant to do. That study's figures reach this site only through a university press release, which is why they are reported at that distance and no page is published for the instigation and execution distinction yet.

In effect

Blood donation against seat belts

The review's own contrast is the clearest research case. For giving blood, voting or having a flu jab, which come round once or twice a year and usually in a different place, stated intention did almost all the predictive work and past behaviour barely registered. For wearing a seat belt, attending class or drinking milk, available every day in the same setting, past behaviour took the larger share. The same two predictors, entered the same way, swap places depending on the behaviour. Any general claim about whether intentions matter has averaged over that swap.

Deciding to go to the gym, and going to the gym

In everyday terms, the moderator explains why a decision feels load-bearing at the start and irrelevant later. A first visit to a new gym needs an intention, because nothing in the setting is doing any work yet. The two-hundredth visit does not, because the route, the hour and the bag by the door are. The honest practical reading is not that intention becomes useless but that its share falls as the setting takes over, which is also why intention-change campaigns bite least where behaviour is already most repeated.

Why a persuasion campaign works on the wrong people

Public-health and marketing campaigns that change what people say they will do are, on this result, best aimed at behaviours that are infrequent and context-unstable, which is where intention carries the prediction. Aimed at daily behaviour in a fixed setting, the same campaign is competing with the half of the mechanism it cannot touch. Neal and colleagues, in 2012, cite a meta-analysis of 47 studies finding that interventions which changed behavioural intentions changed behaviour primarily where behaviour was not habitual, and the vault carries that claim as their description of it rather than as its own read of the underlying paper.

What it does not say

It does not say intention stops predicting habitual behaviour. The weight fell from 0.62 to 0.27, and 0.27 with a confidence interval from 0.20 to 0.34 is not zero.

It does not say habit and intention are rivals. The 1998 path models and the 2025 sampling study both describe them as complementary, and the second finds nearly half of behaviour being both at once.

It is not built on observed behaviour. Past behaviour in most of the reviewed studies was a subjective frequency estimate, which the authors note themselves, so part of what the moderator describes may be a property of how people summarise their own lives.

It is not homogeneous. The 1998 homogeneity test on the pooled matrix was significant, so the studies do not share one population value, and the accompanying primary investigation ran on 141 Texas A&M undergraduates with nine people in the stable-context recycling cell.

Who developed it

Judith Ouellette is first author of the 1998 Psychological Bulletin review, with Wendy Wood second. The review is almost always cited as Wood's; this site credits Ouellette first and treats the result as theirs jointly. The modern synthesis is Martin Hagger's with seven co-authors, and the 2025 sampling study is Amanda Rebar's at the University of South Carolina, with Benjamin Gardner of the University of Surrey fourth author.

---