Contribution
The first meta-analysis of SDT-based experiments in health that tracks both the theory’s constructs and health outcomes, and asks whether changes in the one covary with changes in the other across studies. It extends Gillison et al. 2019, which had motivation outcomes only, and Ng et al. 2012, which was mostly correlational (pp. 216–217).
Key insight
Manipulating need support works on the proximal constructs and, modestly, on behaviour: interventions move perceived support most, need satisfaction and autonomous motivation by about a third of a standard deviation, and health behaviour by about half. What they do not move is relatedness and the controlled side of motivation, and the psychological-health benefit does not outlast the intervention.
Assumptions
- Eligibility: experimental or quasi-experimental; an intervention its authors say was (partly) designed on SDT; at least one SDT construct and one health behaviour, physical or psychological health outcome measured after the intervention. Motivational interviewing without reference to SDT excluded (p. 217).
- Effect sizes: Hedges’ g, random effects, most-versus-least need-supportive arm, baseline-adjusted where possible, last follow-up only; r = .50 imputed to combine correlated outcomes; ICC 0.05 imputed for clustered designs (p. 219).
- “Do changes in SDT constructs engender changes” is tested by regressing one effect size on another across studies (p. 220). This is a between-study association, not mediation within studies, which the authors list as not done (p. 237).
- Searches to November 2018; study selection by one research assistant, not in duplicate (pp. 218, 237).
Key results
The studies (p. 222)
- 73 studies: 68 journal articles, 58 RCTs (20 cluster-randomized), 15 quasi-experimental; 30,088 participants (mean 412), 36.6% male, mean age 35.4; intervention arms had 7.4 (SD 4.6) more SDT strategies than controls; mean duration 133 days; follow-up one week to 30 months.
- Blinding was rare: participants 26.0%, data collectors 11.0%, analysts 8.2%, providers 4.1%.
Effects (Table 1, p. 222)
| outcome | end: k, g [95% CI] | follow-up: k, g [95% CI] |
|---|---|---|
| need support | 21, 0.64 [0.35, 0.93] | 6, 1.13 [−0.35, 2.61] |
| competence | 22, 0.31 [0.12, 0.49] | 11, 0.55 [−0.05, 1.14] |
| autonomy | 17, 0.37 [0.15, 0.60] | 6, 0.25 [−0.01, 0.51] |
| relatedness | 14, 0.20 [−0.04, 0.45] | 6, 0.03 [−0.20, 0.25] |
| combined need satisfaction | 23, 0.37 [0.19, 0.55] | 11, 0.49 [−0.05, 1.02] |
| autonomous motivation | 37, 0.30 [0.17, 0.42] | 14, 0.18 [−0.00, 0.36] |
| controlled motivation | 18, 0.07 [−0.04, 0.18] | 6, 0.02 [−0.24, 0.27] |
| amotivation | 14, −0.07 [−0.28, 0.14] | 5, −0.26 [−0.54, 0.03] |
| health behaviour | 49, 0.45 [0.33, 0.57] | 28, 0.28 [0.17, 0.38] |
| physical health | 16, 0.04 [−0.15, 0.23] | 14, 0.28 [0.03, 0.53] |
| psychological health | 22, 0.29 [0.14, 0.45] | 10, 0.14 [−0.09, 0.36] |
- I² 43–99%. Without outliers (Table 2) competence (0.33), combined need satisfaction (0.28) and autonomous motivation (0.22) become significant at follow-up, and physical health at the end (0.13).
Moderators (pp. 223–232, Tables 3–6)
- Neither the number of SDT techniques nor the difference in techniques between arms moderated any effect (p. 236).
- “Being positive that the individual can succeed” went with larger effects on several constructs and on behaviour at follow-up; its effect on autonomous motivation was confounded with non-specific reward (BCT 10.3), which alone remained significant in a joint meta-regression (B = 0.78) (p. 230).
- Three competence-support techniques (identify barriers, problem solving, develop plans) went with smaller effects on psychological health.
- Allocation concealment reduced effects on autonomous motivation (B = −0.32) and on psychological health at follow-up (−0.74); RCTs had smaller behaviour effects than quasi-experiments (−0.57).
- Exploratory, “hypothesis generating”, no correction for multiple tests (p. 236).
Across-study covariation (Tables 7–8, pp. 228–234)
- With all studies, no SDT effect size predicted the behaviour effect size. After removing one multivariate outlier (Ha et al. 2017), autonomous motivation (β = 0.66) and need support (0.39) did; they also predicted behaviour at follow-up (0.67, 0.35). Need satisfaction effect sizes did not predict behaviour.
- Psychological health effect size covaried with the effect sizes of autonomous motivation (0.58), need satisfaction (0.41), relatedness (0.79), autonomy (0.57), competence (0.39) and need support (0.31), k ≤ 22. Nothing predicted physical health.
- Egger’s test: possible small-study bias for health behaviour (p = .01) and psychological health (p = .006) (p. 230).
Claims
| id | claim | strength | support |
|---|---|---|---|
| C1 | SDT-informed health interventions raise perceived need support, autonomy and competence satisfaction and autonomous motivation | moderate | k = 17–37, end of intervention; rare blinding; heterogeneity high |
| C2 | They improve health behaviour, at the end and at follow-up | moderate | g = 0.45, 0.28; small-study bias possible |
| C3 | They improve psychological health at the end only | weak–moderate | g = 0.29 (k = 22), 0.14 ns at follow-up; small-study bias possible |
| C4 | They improve physical health at follow-up | weak | g = 0.28, k = 14, I² = 92.5% |
| C5 | Gains in need support and autonomous motivation go with gains in behaviour | weak | between-study meta-regression after one outlier is removed |
| C6 | They do not change relatedness, controlled motivation or amotivation | moderate as a null | k = 14–18 |
Concepts
- Need support — behaviour of significant others that supports the three needs, as perceived by the recipient (pp. 215–216).
- Combined need satisfaction — overall or averaged satisfaction of the three needs (p. 219).
- Psychological health — e.g. “quality of life, depression” (p. 220); the breakdown is in Supplementary Table S4, not read.
Connections
- Gillison et al. 2019 found larger effects on autonomy (g = 0.84); the authors attribute the gap to inclusion criteria (p. 234).
- Ng et al. 2012, the correlational health meta-analysis, gave the path model this tests experimentally (p. 216).
- Slemp et al. 2024 (LIT-1969) cites this paper as the experimental counterpart of its correlations.
- The 93-item behaviour change technique taxonomy of Michie et al. 2013 is coded alongside (p. 218).
Bearing on the record
- THEORY-052: the registration overstated what this meta-analysis offers. LIT-1962 was registered as “whether manipulating need support changes outcomes, which THEORY-052’s promote_when asks for”. The promote_when asks for interventions that “reduce need frustration and reduce symptoms of several disorders, mediated by the change in frustration”. This paper measures no need frustration, tests no mediation (p. 237), and pools “psychological health” outcomes in a general population mostly targeted at physical activity. Its psychological health effect is small at the end and gone at follow-up. It does not move THEORY-052.
- It does answer a different question the registration also raised: manipulating need support changes perceived support, need satisfaction and autonomous motivation, and behaviour. That is experimental evidence for SDT’s contextual claim that the correlational syntheses (Slemp et al., NOTE-1507) cannot give.
- QUESTION-369. Two findings bear on need-hood. Relatedness satisfaction did not respond to the interventions (g = 0.20, ns), so the experimental leverage is on autonomy and competence. And the across-study link between need satisfaction and psychological health (β = 0.41) is again a well-being outcome, so it does not escape the interdefinition.
- No dose–response. The number of SDT techniques, and the difference in techniques between arms, predicted nothing (p. 236); a rewards technique outperformed the SDT technique it co-occurred with on autonomous motivation (p. 230). Neither fits a simple story in which more need support gives more benefit.
- THEORY-050 (inflexibility): no overlap; noted because the record pairs the two transdiagnostic accounts.
- Filed from it: THEORY-1320 (Active).
- No instruction for machine-learning practice.
Limitations
- Selection not in duplicate; coders varied; no test of indirect effects (the authors’, p. 237).
- Blinding rare; allocation concealment shrank effects; possible small-study bias on the two outcomes that matter most.
- Most studies are physical-activity promotion; few measured health outcomes or long-term motivation (p. 237).
- Intervention dose (frequency, duration of contact) could not be coded (p. 218).
- Follow-up estimates for the SDT constructs rest on k = 5–14 with very wide intervals.
Open questions
- Does the change in need satisfaction or autonomous motivation mediate the change in behaviour within trials?
- Can an intervention reduce need frustration, and does that reduce symptoms (THEORY-052)?
- Why does relatedness not respond: the techniques used, or the target behaviours (p. 234)?