Contribution
It collects, in one place, the 60 meta-analyses that test hypotheses of self-determination theory (SDT), sorts them by the six mini-theories and by applied domain (Tables 1 and 2), converts their effects to r, and says which propositions have meta-analytic support and which have none. It is a map of the evidence, not a new analysis: no effect is re-pooled, and no included meta-analysis is re-coded or graded for risk of bias.
Key insight
The audit’s own map is uneven, and it says so. Cognitive evaluation theory (CET), organismic integration theory (OIT) and basic psychological needs theory (BPNT) “have received strong meta-analytic support”, while “support is much thinner where COT and GCT are concerned”, and relationships motivation theory (RMT) “has no entry in its column within Table 1” (Summary and Conclusions; Method). Within the supported parts, the support is mostly for associations: “The most striking feature of these meta-analyses, when examined collectively, is their typical (though not invariant) reliance on cross-sectional datasets” (Summary and Conclusions).
Assumptions
- Inclusion. A meta-analysis or systematic review, completed, with SDT variables, testing SDT hypotheses, in English (Method, Study Selection). Records came first “from the personal databases of the paper’s authors” and the Center for SDT website, then from five databases searched in February and July 2021. Two reviewers screened; 33 full texts were excluded (Table S3, not read); 60 were kept, 58 of them published.
- No re-analysis. Effects were extracted and converted to r (d to r with a = 4, after Borenstein et al.); k, N, I², τ² and Q are listed in Table S1 where reported (Data Collection).
- “This review of meta-analyses was not preregistered” (Transparency and Openness).
- Effect-size benchmarks. It reads effects against Gignac and Szodorai’s percentiles (r = .11, .19, .29) and Funder and Ozer’s (r = .10 small, .20 medium, .30 large), not Cohen’s (Why Meta-Analyses?).
- Authorship. The first author is SDT’s co-founder; several included meta-analyses are by the authors (asterisked references include Donald et al. 2020, 2021; Bradshaw et al. 2021, 2022; Duineveld 2018; Gillison et al. 2019, with Ryan). The review does not discuss this.
Key results
All r; k = studies or samples as the review gives them. Sections are the manuscript’s.
CET
- Choice raises intrinsic motivation, r = .15 (Patall et al. 2008; 41 studies, 46 samples); “instructionally irrelevant” choices had the largest effects, which “CET does not offer a more specific explanation for”.
- Negative vs positive feedback, r = −.18 (Fong et al. 2019; 78 studies). Negative vs no or neutral feedback: no significant difference, heterogeneous; in the 15 published studies negative feedback raised intrinsic motivation, in the 30 unpublished it lowered it.
- Rewards (Deci et al. 1999, re-analysing Eisenberger and Cameron’s data): all rewards −.12; verbal .16; tangible −.17; unexpected tangible .00; expected −.18; task-noncontingent −.07 (n.s.); engagement-contingent −.20; completion-contingent −.18; performance-contingent −.14.
- Intrinsic motivation and performance (Cerasoli et al. 2014; k = 183), r = .21; with incentives directly tied to performance .21, indirectly .34. Unique prediction of quality: intrinsic β = .35, incentives .06; of quantity: incentives .33, intrinsic .24.
OIT
- The continuum. Chatzisarantis et al. (2003; 21 articles, sport) found a simplex. Howard et al. (2017; 486 samples, N > 200,000) “largely supported a continuum-like structure”, but “due in part to both too few studies measuring integration and its high correlation with identified regulation, results did not support the differentiation of integrated regulation from identified regulation”, nor Vallerand’s three intrinsic subtypes (Testing the Simplex Model). Van den Broeck et al. (2021; 124 work samples): integrated regulation “was rarely measured … and where it was measured it did not show differentiation from identified motivation” (Work and Organizations).
- Graded relations with support. Lochbaum and Jean-Noel (2016): autonomy-supportive PE teaching with intrinsic .54, identified .50, introjected .20, external −.15, amotivation −.19. Mossman et al. (2022; 131 samples): coach autonomy support with integrated .37, intrinsic .32, identified .31, introjected .13, external .00, amotivation −.16.
- Rationale provision (Steingut et al. 2017; 23 experiments): task value .16, perceived autonomy .20, engagement .10, performance .08, and perceived competence −.10.
- Outcomes (Howard et al. 2021; 344 samples, N = 223,209, education): intrinsic and identified with performance and well-being; introjected weakly with persistence and negatively with well-being; amotivation worst. Osbaldiston (2005, unpublished) found OIT categories did not predict pro-environmental outcomes.
- Prosociality (Donald et al. 2021; 138 studies): autonomy with prosocial outcomes .28, control with antisocial .16; the longitudinal effect of control “non-different from zero”; one intervention study.
COT
- Hagger and Hamilton (2021; 69 studies, 83 samples): the autonomy orientation correlates most with intrinsic and identified regulation, the control orientation with external and introjected; correlations with outcomes “small-to-medium”. Impersonal–control r = .27, called “large”.
- Murphy and Steel (2021; 30 priming experiments, 33 effects): evidential value survives removal of the largest effects.
- The section opens “only two meta-analyses”; its summary says “Only one meta-analysis has thus far examined this proposition”.
BPNT
- Need satisfaction with positive affect .39 to .45, with negative affect −.30 to −.33 (Stanley et al. 2021; 16 studies), moderated by gender, sample and instrument.
- Autonomy with subjective well-being .46, not moderated by East Asian vs Western culture (Yu et al. 2018; 36 samples, about 13,000).
- Slemp et al. (2018; 83 samples, N = 32,870): leader autonomy support with autonomy .46, competence .34, relatedness .38 satisfaction; not moderated by country.
- “The proposition that frustration of these needs is associated with ill-being has yet to be meta-analytically investigated” (BPNT Summary). Van den Broeck et al. (2016): needs predicted positive outcomes more robustly than negative ones, and its authors “questioned if they should be considered truly basic needs”.
GCT
- Bradshaw et al. (2022; 92 studies, 70,110 people, 27 countries): intrinsic aspirations with well-being .24, ill-being −.11; extrinsic aspirations with well-being .02 overall. Split by scoring: as a simple score .07 with well-being; as a relative-centrality score −.22 with well-being and .23 with ill-being. Not moderated by gender, age, region or SES.
- Dittmar et al. (2014; 259 samples): materialism with life satisfaction −.13, positive affect −.23, compulsive buying −.44; need satisfaction mediates. “the mediational effect of need satisfactions has yet to be meta-analytically established for intrinsic aspirations”.
RMT and parenting
- No meta-analysis of RMT. Koehn and Kerns (2018; 10 studies, not SDT-framed): parental autonomy support with child attachment security .18; harsh control −.20. Bradshaw et al. (2021, under review; 211 reports): parental autonomy support with child well-being .28, ill-being −.22; control with ill-being .19, well-being −.12. Duineveld et al. (2018): autonomy support and psychological control r = −.50 overall, .10 in early childhood, −.72 in emerging adulthood.
Interventions
- SDT-based health interventions on behaviour, r = .11 (Sheeran et al. 2020; 65 tests); autonomy-support training effective, r = .30 (Su and Reeve 2011); need-support and autonomous motivation changes, not controlled motivation, track behaviour change (Ntoumanis et al. 2021).
Claims
| id | claim | strength | support |
|---|---|---|---|
| C1 | Meta-analytic support for SDT is concentrated in CET, OIT and BPNT; COT and GCT have one or two meta-analyses each; RMT has none | strong (as a description of the 60) | Tables 1, 2; Summary and Conclusions |
| C2 | The relative-autonomy continuum is supported as an ordering of motives | moderate | Howard et al. 2017 and Chatzisarantis et al. 2003 as reported; simplex correlations, all self-report |
| C3 | Integrated regulation has not been separated from identified regulation | strong (as reported, two meta-analyses agree) | Howard et al. 2017; Van den Broeck et al. 2021 |
| C4 | Need satisfaction correlates with well-being, invariantly across individualist and collectivist samples | moderate | Yu et al. 2018; Slemp et al. 2018, 2020; correlational |
| C5 | The link of need frustration to ill-being has no meta-analysis | strong (as a statement about the literature to July 2021) | BPNT Summary |
| C6 | Relative centrality of extrinsic aspirations correlates with lower well-being; simple extrinsic scores do not | moderate | Bradshaw et al. 2022, scoring as moderator |
| C7 | The evidence base is mostly cross-sectional and heterogeneous; causal support rests on the experimental and intervention meta-analyses | strong (the authors’ own judgment) | Summary and Conclusions; Heterogeneity |
Concepts
- Meta-review — here, a narrative synthesis of published meta-analyses with their effects tabulated, not a second-order pooling.
- Relative centrality score — an aspiration’s importance minus, or controlling for, the mean importance of all aspirations; the “simple score” is the raw mean, which the authors read as carrying “a ‘general striving’ component” (GCT).
- Dual-process model — need support does more for well-being than against ill-being, and control more for ill-being than against well-being (Parenting).
- Six mini-theories — CET, OIT, COT, BPNT, GCT, RMT, each with formal propositions listed in Table S2 (not read).
Connections
It is the record’s first second-order look at SDT’s evidence. Of the record’s own SDT readings, it re-reports Deci, Koestner and Ryan 1999 (LIT-557) with the same numbers as NOTE-438, and it cites Ryan, Mims and Koestner 1983 (LIT-1970) and Deci and Ryan 1985 (LIT-1955) for the reward taxonomy and COT. It does not cite Cameron, Banko and Pierce 2001 (LIT-562): the reward dispute is told through Cameron and Pierce 1994, Eisenberger and Cameron 1996 and Eisenberger et al. 1999, the last of which, it says, “did not dispute the errors discovered”.
Bearing on the record
- THEORY-054: confirms what A2 said of Howard et al. 2017, and adds a second source. The review reports that Howard et al. did not differentiate integrated from identified regulation, partly because too few studies measured integration, and that Van den Broeck et al. (2021) found the same at work. So THEORY-054’s first promote_when clause, a measure of integration separable from identification, is unmet in the meta-analytic literature as of 2021. This is second-hand; Howard et al. (LIT-1996) is not read. Mossman et al.’s r = .37 for integrated regulation with coach autonomy support is not evidence of separability, since it is no different in kind from identified (.31).
- THEORY-052: the review says the needed meta-analysis does not exist. “The proposition that frustration of these needs is associated with ill-being has yet to be meta-analytically investigated” (BPNT Summary). THEORY-052 is about frustration as a transdiagnostic process; the review’s satisfaction–well-being correlations do not bear on its promote_when, which asks for temporal precedence measured without shared self-report content.
- QUESTION-369: the review’s evidence is of the kind the question says cannot settle it. Needs are supported by correlations of need satisfaction with well-being and affect (Stanley et al.; Tang et al.; Yu et al.). Van den Broeck et al.’s doubt that needs predicting positive more than negative outcomes are “truly basic” is the nearest the review comes to the question, and the authors answer it with frustration measures, which have no meta-analysis.
- THEORY-049 and THEORY-154. The review states CET’s third proposition as a mediation claim, that “experiences of autonomy versus control, and of competence versus lack of it, mediate the relations between external events and their … effects”, and reports no meta-analytic test of that mediation. THEORY-049’s promote_when stands unmet. The review does not engage LIT-562.
- THEORY-047. Steingut et al.’s rationale meta-analysis (23 experiments) is the nearest meta-analytic evidence to Deci et al. 1994 (LIT-1973): a rationale raises perceived autonomy (.20) and task value (.16), and is “most effective” when delivered autonomously. It measures present endorsement, not the history-versus-structure contrast THEORY-047’s promote_when asks for.
- RMT, the registration’s target for LIT-1979 and LIT-1971. The review confirms A1’s table: no RMT meta-analysis exists.
- QUESTION-066. No meta-analysis in the review tests persistence after support is removed with valuation held fixed; Howard et al. 2021’s persistence outcomes are cross-sectional.
- No instruction for machine-learning practice.
Limitations
- Not independent. The lead author founded the theory; several included meta-analyses are the authors’ own. The registration’s “the nearest thing to an audit” holds only as a map of the evidence; nothing in the review grades the quality of the meta-analyses or their primary studies.
- Not preregistered, narrative, English only, and seeded from the authors’ own databases.
- Its own counts disagree. COT has “only two meta-analyses” in one paragraph and “only one” in its summary; GCT has “only one primary meta-analysis” in the conclusions after two (Bradshaw et al. 2022, Dittmar et al. 2014) were reviewed.
- A slip in the GCT section. “When extrinsic aspirations were calculated as a simple score their link to well-being was positive but very small (r = .07)” is repeated where the sentence is about ill-being. The accepted manuscript only; the published text was not checked.
- “Large” for r = .27 between the impersonal and control orientations, against its own benchmark (Funder and Ozer’s .30 is large).
- Heterogeneity is reported, not explained: “accounts of heterogeneity have been far from comprehensive” (Heterogeneity).
- The supplement (Table S1’s k, N, I² per effect) was not read, so no effect above was checked against its source meta-analysis.
Open questions
- Is integrated regulation measurable apart from identified regulation at all, or is it a theoretical category with no indicator?
- What would a meta-analysis of need frustration and ill-being, with frustration measured from contexts rather than self-report, find?
- Would an independent risk-of-bias assessment of the 60, especially of the correlational OIT and BPNT literature, change the “meta-analytically true” list?