The decision this page changes. Someone who already intends to exercise often does not. The implementation intention (Gollwitzer) is a written if-then plan for when, where and how the activity will happen, sometimes with a second coping plan for likely obstacles. It costs minutes and has no harm. The question is whether it adds activity beyond what motivation alone produces, so that it belongs on the action list for a new exerciser. Like a tracker, it does not change what a dose of activity does (Physical Activity Dose and Mortality); it is a candidate tool for reaching and keeping the dose (an intervention not done has no effect, so adherence is part of the effect). (inferred from Silva et al., 2018)
Verdict — no significant pooled effect; trials that revisited the plan show a small self-reported effect against their own controls
Across 11 RCTs the pooled effect of planning on self-reported activity is small and not statistically significant (SMD 0.15, 95% CI -0.01 to 0.31). The trials that reinforced the plan during follow-up (phone call, meeting or text) show SMD 0.25 (0.05-0.45); the four that did not show essentially zero. Every outcome is self-reported, follow-up runs 2-32 weeks, and the reinforcement contrast is a between-trial comparison that the review did not formally test. On this evidence, a plan made once and left alone is not shown to work; a plan that someone checks back on may add a little, but whether it beats a one-off plan is untested. (inferred from Silva et al., 2018)
The pooled evidence (Silva 2018 SR+MA)
Silva et al. pooled RCTs to September 2016 of implementation-intention strategies for physical activity in adults: 13 trials reviewed, 11 meta-analysed (12 comparisons), random-effects SMD, PROSPERO-registered. Populations were mostly clinical or at-risk (77%: spinal-cord injury, coronary disease, obesity, diabetes, sedentarism, occupational stress); ages 18-76; trial sizes 30-709. (Silva et al., 2018)
- Overall: not significant. «In the general analysis, when the summary of the effect was analyzed, the result shown was 0.15 (IC 95% = -0.01–0.31) demon- strating that the implementation intentions strategy did not significantly increase the practice of PA in comparison with the groups that did not receive this strategy.» I2 = 54%. (Silva et al., 2018)
- Reinforced trials: SMD 0.25 (0.05-0.45), 7 trials / 8 comparisons, I2 = 47%, 61.3% of the weight. Non-reinforced trials: SMD -0.01 (-0.27 to 0.24), 4 trials, I2 = 59%. The non-reinforced estimate is printed only on the forest plot (Fig 2), not in the text. (Silva et al., 2018)
- Certainty is not graded. Quality was scored on the Jadad scale (12 of 13 trials at 3 of 5); no GRADE rating, no risk-of-bias tool beyond Jadad. Publication bias was not tested: «Due to this subdivision of studies, the publication bias was not evaluated, as there were not enough studies to be grouped into a funnel plot.» (Silva et al., 2018)
- The comparator is motivation, not nothing. «the majority of the studies (92%) applied a motivational phase to their participants» (videos, pamphlets, meetings). So the estimate is the increment of planning over motivational information. (Silva et al., 2018) This makes a near-null estimate a statement about the added value of a plan, not evidence that the whole package does nothing. (inferred from Silva et al., 2018)
What the reinforcement subgroup does and does not show
The headline the authors carry — reinforcement works — rests on the subgroup, and the subgroup is a weaker object than its CI suggests. (inferred from Silva et al., 2018)
It is not a test of reinforcement. The 0.25 is the reinforced trials’ plan arms against their own
controls. No test of the difference between the two subgroups is reported in the text or on Figure 2
[searched: subgroup differences / between subgroups / interaction / test for across chunk 01 — 0 hits].
(Silva et al., 2018) An approximate
z-test from the two printed CIs (difference 0.26, SE ~0.17) gives p ~0.12, so the subgroup contrast is
itself not significant. That figure is the wiki’s arithmetic, not the source’s.
(inferred from Silva et al., 2018)
Reinforcement is a trial feature, not a randomized factor. All the coronary-disease, spinal-cord-injury and diabetes trials sit in the reinforced subgroup; obesity, sedentary and healthy-adult trials appear in both subgroups. Reinforced trials also differ in delivery (phone, meeting, text). (Silva et al., 2018) Treating reinforcement as an effect modifier needs positive interaction evidence (route (b)); this is a between-trial association. (inferred from Silva et al., 2018)
One shared control may be counted twice. On Figure 2 both Prestwich 2009 comparisons print an identical control arm (n 65, mean 0.25, SD 1.22), and the reinforced subgroup’s control total (490) sums both. The text describes two control groups, and also two intervention groups: «one that worked with the strategy of implementation intentions and the other with the implementation inten- tions associated with text messages». Both comparisons are pooled as reinforced, and the review does not say which row carries which arm. (Silva et al., 2018) If the control is one group, the subgroup’s precision is overstated; and one of the eight reinforced comparisons appears to test a plan with no text-message reinforcement. (inferred from Silva et al., 2018)
The coping-planning claim is narrative. The authors attribute the null trials to missing coping plans: «With regard to the four studies [30–33] that presented no significant differences in the group that received the strategy of implementation intentions, the absence of the use of coping planning, was factor in common among them, thus demonstrating the importance of obstacle management until the behavior has indeed been transformed into habit among the partici- pants.» (Silva et al., 2018) No coping-planning subgroup was pooled. On Figure 2 the two individually significant trials (Luszczynska 2006, 0.53; Prestwich 2009 a1, 0.48) used action planning without coping planning, and all four meta-analysed coping-planning trials have CIs crossing zero. (Silva et al., 2018) The coping-planning reading is not supported by the review’s own forest plot. (inferred from Silva et al., 2018)
The conclusion outruns the estimate. The authors conclude that planning «seemed to promote PA behavior in different populations and countries, particularly when there was incorporation of plans for management of perceived obstacles», and that reinforcement «appeared to be effective both for engaging in and maintaining PA practice». (Silva et al., 2018) Against that: the primary pooled estimate is null, the coping-planning contrast was not pooled, and no trial ran past 32 weeks. (inferred from Silva et al., 2018)
Reviewer-trial overlap. Two of the 11 meta-analysed trials each list a review author (Silva 2015, SMD 0.67; Rodrigues 2013, SMD 0.21), together 13% of the weight. Both are reinforced trials. The authors declare no competing interests. (Silva et al., 2018) The two trials carry about a fifth of the reinforced subgroup’s weight (13.2 of 61.3 points). (inferred from Silva et al., 2018)
Measurement — every outcome is self-report
- «Relative to the type of instrument used for measuring PA practice, 100% of the studies used self-report instruments to assess behavior.» Over half (54%) used non-validated duration/frequency questions. (Silva et al., 2018)
- The authors name the bias and its direction: «In this sense, social desirability is understood as the tendency of respondents to distort self-reports in a favorable direction, and this might compromise self- reports on PA, by overestimating them [43].» (Silva et al., 2018)
- Why it matters here more than usual. A participant who has just written a plan, or been phoned about it, has more reason to report having followed it than a control. Differential reporting of this kind could produce a reinforced-arm effect with no change in movement. It is untestable without a device-measured outcome, and no trial had one. (inferred from Silva et al., 2018)
- No minutes or steps. The review gives no conversion of its SMD into activity units
[searched: steps / minutes across chunk 01 — steps 0 hits; minutes only in WHO-guideline background]. An SMD is relative to the spread of the instrument that produced it, so a 0.25 here cannot be read against the tracker page’s step-measured SMD 0.23 (~627 steps/day): different instruments, different denominators. (inferred from Silva et al., 2018)
Durability — short horizons, and guidance says maintenance is the weak point
Silva’s trials ran «from a minimum of two to a maximum of thirty-two weeks». (Silva et al., 2018) The ACSM 2011 position stand states the general pattern for behavioural PA interventions: «Individually tailored interventions deliv- ered using various modalities including print (43), telephone, Internet/computer (374), and group counseling (303) can be effective in enhancing exercise adoption but are at best marginally effective for increasing exercise maintenance.» and «few data exist regarding the factors associated with maintaining exercise behavior». It adds: «Successful interventions to improve exercise main- tenance have incorporated continued contact and social sup- port (68) and access to home exercise equipment (182).» (Garber et al., 2011)
| Parameter | Silva 2018 | Garber 2011 (ACSM) | Same quantity? |
|---|---|---|---|
| Intervention | implementation intentions (if-then planning) | individually tailored behavioural interventions, many techniques | NO — Garber’s class is broader |
| Outcome | self-reported PA, SMD, 2-32 weeks | narrative: adoption vs maintenance | NO — no pooled estimate in Garber |
| Durability statement | none past 32 wk; authors claim maintaining from short trials | «at best marginally effective» for maintenance | NO — different objects; direction only |
(Garber et al., 2011; inferred from Silva et al., 2018)
The two sources do not measure the same thing, so this is no independent confirmation. What they share is a gap: Silva’s trials stop at 32 weeks (about 7 months; the Discussion says 34), and Garber names ingredients of successful maintenance interventions (continued contact among them) without a horizon or a magnitude. Garber’s continued contact points the same way as Silva’s reinforcement subgroup, but it is a narrative statement about a broader intervention class, not a test. Silva’s maintenance wording is not backed by its own follow-up. (Garber et al., 2011; inferred from Silva et al., 2018) (type-G)
How to use it
- Do it anyway, as a cheap default — labelled as such. Writing down when, where and how the next sessions will happen, and what to do when the usual obstacle appears, takes minutes, costs nothing and has no plausible harm. The review does not show it works on its own. In the trials that came back to the plan (a call, a meeting, a text), planners reported more activity than their own controls (SMD 0.25); that this beats a one-off plan is untested. So the default is a plan plus a scheduled check-in to revise it, not a plan filed and forgotten. Basis: low-cost practice consistent with a self-reported, between-trial signal — not a trial-tested effect. The trial check-ins were researcher contact; a self-scheduled review is a different, untested exposure, and the extra attention (or simply being asked about activity) may be the active part rather than the plan. (inferred from Silva et al., 2018)
- Make the plan revisable. The review cites Latimer’s proposal that plans be updated at each reinforcement, and Budden and Sagarin attributing their null to «inflexibility of the plans executed». (Silva et al., 2018)
- Do not rank it as a lever in its own right. Expected effect is small (point estimate SMD 0.25, CI 0.05-0.45, in the most favourable subgroup, on self-report) and of unknown duration. It is an aid to pulling the inactivity rock, not a rock. (inferred from Silva et al., 2018)
Gaps
- Device-measured outcome. No trial measured activity with an accelerometer or pedometer, so the effect on actual movement is unknown (type-G). (inferred from Silva et al., 2018)
- A randomized test of reinforcement (plan vs plan + scheduled check-ins, same population) would settle whether the check-in is the active ingredient (type-G). (inferred from Silva et al., 2018)
- The earlier, larger MA. — resolves whether a positive pooled effect holds on a different trial set and whether any device-measured subset exists; Silva cites it as finding a positive effect, which this page does not restate.
- Durability past 32 weeks (see above). (Garber et al., 2011; inferred from Silva et al., 2018)
Synthesis — check-ins and prompts recur as the moderator across adherence tools
Silva’s untested reinforcement subgroup, Laranjo’s exploratory text-messaging moderator and Brickwood’s narrative framing all point toward repeated contact or prompting. Each is at most a between-trial association, and they measure different things. (inferred from Silva et al., 2018)
| Parameter | Silva 2018 (planning) | Laranjo 2020 (apps/trackers) | Brickwood 2019 (trackers) | Same quantity? |
|---|---|---|---|---|
| Contrast | reinforced vs non-reinforced trials, SMD 0.25 vs -0.01, no difference test | text-messaging trials SDM 0.495; meta-regression 0.422 | authors’ framing: ongoing professional contact sustains adherence | NO — three different moderator constructs |
| Outcome | self-report PA | mostly device-measured (3 of 28 trials self-report) | steps (objective) and others | NO |
| Status | between-trial, untested difference | exploratory per its authors | narrative framing | NO |
(inferred from Silva et al., 2018)
The nearest planning signal on a device-measured outcome is Laranjo’s: app and tracker trials whose interventions carried goals and planning behaviour-change techniques showed SDM 0.446 (0.33-0.562) as a subgroup, but in the first multivariate meta-regression model (the significant subgroup moderators entered together, text messaging among them), the goals-and-planning coefficient was 0.05 (-0.041 to 0.141, p 0.281) while text messaging stayed at 0.365 (0.107-0.624). Laranjo describes its meta-regression as exploratory. (Laranjo et al., 2020) A BCT category coded inside app trials is not the same quantity as a written if-then plan, so this is a related construct, not a replication. It leans the same way as Silva’s subgroup (the prompt, not the plan), on an outcome that is mostly device-measured. (Laranjo et al., 2020; inferred from Silva et al., 2018)
None of Silva’s 13 trials appears in Brickwood’s or Laranjo’s pools, so the overlap is not shared data. But no row is the same quantity, so this is a question, not a finding: is the active ingredient of these adherence tools the repeated prompt rather than the tool? It stays a hypothesis until a trial randomizes the prompt itself (type-C, provisional). (inferred from Silva et al., 2018)