Implementation intentions and the if–then form

Peter Gollwitzer, a psychologist working at the University of Konstanz and later at New York University, drew a distinction in the 1990s between two kinds of mental commitment. A goal intention names a desired end state. An implementation intention names a situation and a response, joined in a conditional form: if a particular cue occurs, then a particular action follows. Gollwitzer's argument was that the second form does something the first does not — it delegates the initiation of the action to the environment rather than leaving it to be decided in the moment.

The mechanism proposed was that the specified cue becomes more accessible in memory, and that the link between cue and response acquires some of the automaticity ordinarily associated with habit. The claim is narrow. It concerns the initiation of an action that has already been chosen, not the motivation to choose it.

The meta-analysis

Gollwitzer and Paschal Sheeran published a meta-analysis in 2006 covering roughly ninety independent tests across health behaviour, academic performance and laboratory tasks. The pooled estimate was in the medium-to-large range by conventional standards — unusually large for a manipulation that consists of asking participants to write a sentence.

Individual studies in that set are worth reading directly rather than in summary. Sarah Milne, Sheila Orbell and Sheeran ran an exercise study in which one group received motivational material and another additionally formed a written if–then plan specifying when and where activity would take place. The difference in participation rates between those groups was large. The study is small by present standards and was not pre-registered.

Where the estimate moved

Two qualifications matter. The first is publication bias. The literature is composed largely of small samples, which is the condition under which selective publication inflates pooled estimates most severely. Later analyses applying bias-correction methods have returned smaller figures than the 2006 paper, and researchers within the programme have acknowledged the pattern rather than disputing it.

The second is heterogeneity across populations and time horizons. Ilenia Toli, Thomas Webb and Gillian Hardy examined the effect in clinical populations and reported estimates below those found in general samples. Reviews by Martin Hagger and Aleksandra Luszczynska have noted that effects observed over weeks frequently do not persist over months, and that outcomes vary considerably with how the plan was elicited and whether the underlying goal was one the participant already held.

That last condition is the one most often dropped in secondary accounts. Across the literature the manipulation shows little effect where the participant is indifferent to the goal. It operates on the gap between an existing intention and its execution, which is a narrower target than the popular framing suggests.

Where the record stands

Cairn Post marks this entry as supported. Independent laboratories have reproduced the pattern across varied domains, the theoretical account has been specified precisely enough to be tested, and the researchers involved have engaged with the bias critique rather than deflecting it. The honest summary is that the direction of the effect is reasonably secure and its magnitude is smaller than the earliest pooled estimates indicated.

Readers who want to evaluate this themselves should begin with the 2006 meta-analysis and then read a bias-corrected re-analysis alongside it, in that order. The difference between the two figures is the most informative thing in the file.