Contribution
Before this chapter the effect of if-then planning had been reviewed only narratively or in small quantitative reviews (p. 70). It gives the first large meta-analytic estimate (94 tests), sorts the tests by four problems of goal striving (getting started, getting derailed, not calling a halt, overextending oneself), and separately pools the tests of the two postulated processes: heightened accessibility of the planned cue and automatic running of the planned response.
Key insight
A goal intention says what one wants; an implementation intention says in advance which situation will trigger which response. Forming one “involves a strategic abdication of effortful action control” (p. 84): “forming an implementation intention means deciding these issues in advance, thereby delegating the control of goal-directed behavior to specified situational cues” (p. 100). The authors’ own phrase for the result is “instant habits” (p. 84). The plan is a pre-decided mapping from cue to response, and its benefit is that nothing has to be deliberated in the situation.
Assumptions
- The comparison is plan plus goal against goal alone. d is the difference between participants who formed a goal intention and an implementation intention and those who formed only the goal intention (pp. 92, 98).
- Inclusion. A plan specifying a goal-directed response on an internal or external cue, and a retrievable statistic; searches January 1990 to December 2003, with unpublished studies requested from authors (p. 87). Eleven of the 94 tests were correlational (Table II, p. 93).
- Pooling. Hedges and Olkin’s bias-corrected d, inverse-variance weighted; non-independent samples averaged and given the smallest N (p. 92). Fixed-effect weighting with Q tests of homogeneity; overall Q(93) = 173.46, p < .001 (p. 93).
- The sample. Of 94 tests, 79 were with university students; 38 were laboratory tasks and 23 health goals (Tables II and III, pp. 93, 95).
Key results
- Overall (p. 92). d = .65, 95% CI [.60, .70], k = 94, N = 8,461.
- Moderators (pp. 93–94, Table II). No difference between correlational and experimental designs (.70 and .65), self-report and objective outcomes (.63 and .67), or published and unpublished tests (.65 and .67). Three tests with people with psychological problems (schizophrenia, frontal-lobe injury, heroin withdrawal) gave d = 1.10 against .66 for the rest, Q(1) = 4.54, p < .04.
- Goal domains (Table III, p. 95). Medium or large in all eight, from consumer (.41, k = 2) to environmental (1.12, k = 3).
- Self-regulatory problems (Table IV, p. 96). Getting started d = .61 (k = 52); shielding from unwanted influences .77 (k = 21); disengaging from failing goals .47 (k = 3); conserving capability for a later task 1.28 (k = 3), meaning no ego-depletion or stereotype rebound after the planned first task (p. 97).
- Component processes (Table V, p. 97). If-component (cue detection, discrimination, accessibility, attention, memory) d = .80, k = 6, three of the five measures from a single test each. Then-component: immediacy .77 (k = 7), efficiency under cognitive load .85 (k = 7), no need of conscious intent .72 (k = 3).
- Not a motivational effect (pp. 100–101). Studies measuring goal intention and self-efficacy before and after planning found no increase; planning still worked when both were near ceiling before it.
- Who benefits most (p. 101). “forming implementation intentions is most likely to benefit goal achievement when regulating the behavior is difficult or people have chronic difficulties in regulating their behavior.”
- Rigidity (pp. 103–104). The authors take up whether plans make goal striving rigid and answer no on three counts, from narrative evidence, not pooled tests: plan effects appear only when the goal intention is strong and activated, so “action control by implementation intentions does not involve a mechanistic elicitation of action in the presence of environmental cues but rather respects the presence versus absence of activated strong goal intentions” (p. 103); speeding one aspect of performance did not cost accuracy or variety; and planners are argued, not shown, to stay sensitive to alternative opportunities. A fourth question is left open: what people do when a plan fails. “We do not know yet what happens” then (p. 104).
- Plans can be bad (p. 102). They help only if precise, viable and instrumental, and “How often” everyday plans meet these conditions “is an empirical issue”.
- Mediation (p. 106). Only two studies had tested mediation: cue accessibility (Aarts et al. 1999), and cue accessibility with the strength of the cue–response link (Webb & Sheeran). Mediation by immediacy, efficiency and lack of intent was untested.
- Repetition makes habits (p. 108). Acting repeatedly on a plan produced more reported features of habit than acting on a goal intention (Orbell & Verplanken 2005, cited).
Claims
| id | claim | strength | support |
|---|---|---|---|
| C1 | Adding an if-then plan to a goal intention raises goal attainment, d = .65 | moderate as a size: no correction for publication bias beyond a published/unpublished split; the 2024 review of 642 tests puts it far lower | 94 tests, p. 92 |
| C2 | The effect holds for starting, shielding, disengaging and conserving capability | strong for starting (k = 52), weak for disengaging and conserving (k = 3 each) | Table IV, p. 96 |
| C3 | Planned responses run immediately, efficiently and without conscious intent | moderate: large pooled effects from 3–7 tests each | Table V, p. 97 |
| C4 | Planned cues become more accessible, and accessibility mediates goal attainment | moderate: k = 6 for the effect, two studies for mediation | pp. 97, 106 |
| C5 | Plans work by delegating control to cues, not by raising motivation | moderate | pp. 100–101 |
| C6 | Plan effects depend on an active, strong goal intention, so plans do not fire mechanically | weak to moderate: narrative, a handful of studies | p. 103 |
| C7 | Plans do not make goal striving rigid | weak: argued; the case of a failing plan is untested | pp. 103–104 |
Concepts
- Goal intention: “I intend to reach Z!”; the end of deliberation about what to achieve (p. 70).
- Implementation intention: “If situation Y is encountered, then I will initiate goal-directed behavior X” (p. 69). The then-part may also be suppressing a response or ignoring a stimulus (p. 83).
- Intention–behaviour gap: intenders who do not act; the median share was 47% in Sheeran’s 2002 review (pp. 72–73).
- Model of action phases: predecisional, preactional, actional and postactional phases, after Heckhausen and Gollwitzer (pp. 74–75).
- Instant habit: the authors’ term for the cue-triggered, automatic responding a plan creates without repetition (p. 84).
Connections
It rests on Gollwitzer’s theory of implementation intentions (1993, 1999) and the model of action phases, and on Bargh’s features of automaticity. Its successor is Sheeran, Listrom & Gollwitzer’s meta-analysis of 642 tests (LIT-1488, read in NOTE-1131), which finds substantial publication bias and a much smaller corrected effect. The “if…then” profiles of Mischel and Shoda (1995) are named as personality’s own situation–behaviour links, which a plan could be tailored to (p. 107).
Bearing on the record
- CLAIM-443 (a learned prior over which processes govern). The
claim cites this work as showing that implementation intentions are
pre-arranged if–then links of situation to response. That is right, and
the reading sharpens what kind of link it is.
- It is a prior over actions, in the claim’s terms. The plan fixes the response to a cue in advance; its benefit is that the person no longer “has to deliberate in situ about what goal-directed response to undertake” (p. 100). That is the cached action policy the claim contrasts with a learned relation of influence, not an example of the latter. Filed as THEORY-913.
- Its gating is goal-dependent, which the claim does not mention. Plan effects appear only when the goal intention behind them is active and strong (p. 103). So the cached mapping is not unconditional: a higher-level state still decides whether it fires. That is a third arrangement beside the claim’s two (cue-to-action, and recruitment of a process that chooses): a cue-to-action mapping licensed by a goal. Filed as THEORY-904.
- The claim’s test is the authors’ open question. The claim distinguishes the two priors by what happens “when the appropriate action changes”. The authors raise exactly this case, a plan that “failed to lead to goal attainment (or even produced negative outcomes)”, and say it has not been studied (p. 104). Twenty years later the 2024 review does not report it either.
- Some evidence of saving, not only persistence. The claim says persistence is not amortization unless earlier work replaces later computation. The efficiency tests (planned responses unaffected by a concurrent load, k = 7) and the conserved-capability tests (no depletion or rebound afterwards, k = 3) are measured savings of later effort, not just stability (pp. 85, 96–97). They are small sets, and they show saving for a cached action, not for a learned allocation of authority.
- CLAIM-358 (learned accommodation reduces amortized cost). The same efficiency and capability results are the nearest the record holds to a measured amortization in people: a decision taken once, in advance, costs less at each later encounter. They do not involve negotiation among processes.
- CLAIM-349 (access, authority and action are separable). The ignore responses (p. 83; Appendix I) cut a stimulus’s access to action by plan; the goal-dependence result keeps the plan’s own authority conditional. A loose parallel only.
- CLAIM-277 (durable authority from endorsement, and its entrenchment). Repetition of planned actions produced reported habit (p. 108), the route by which a planned mapping could entrench. Cited, not tested here.
- QUESTION-095 (caching computation). The plan is a cached decision in the plain sense; this chapter’s efficiency results bear on it.
- No instruction for machine-learning practice; nothing for the anthology.
Limitations
- Publication bias. Checked only by a published/unpublished split (p. 94), the basis for the authors’ view that the review is unlikely to suffer from a file-drawer problem (p. 98). No funnel plot or trim and fill. The 2024 review of 642 tests finds Egger’s b = 1.06, “substantial”, and a robust Bayesian estimate of d = .15 (NOTE-1131, THEORY-918).
- Fixed-effect pooling with heterogeneous effects (Q(93) = 173.46). The confidence interval [.60, .70] is narrower than a random-effects one would be.
- Mostly students and laboratory tasks; 11 correlational tests in the pool.
- The process evidence is thin. Several component estimates rest on one test each; mediation was tested in two studies, for accessibility and link strength only.
- The rigidity defence is narrative, and the failing-plan case is untested by the authors’ own account.
- Many tests are by the authors’ own groups (Table I), which the chapter does not discuss.
- OCR copy. Numbers were read from an OCR of a scan; a few were checked against table alignment by eye.
Open questions
- What happens to a planned mapping when the planned response becomes the wrong one: is it dropped, revised or kept? A design that trains a plan, then changes which response is correct, would answer the authors’ question and test CLAIM-443’s distinction at once.
- Do immediacy, efficiency and lack of intent mediate goal attainment, as accessibility does?
- Does goal-dependence survive repetition, or does a repeated plan become a habit that fires without the goal?