Source
Ntoumanis, Ng, Prestwich, Quested, Hancox, Thøgersen-Ntoumani, Deci, Ryan, Lonsdale and Williams (2020), LIT-1962, pp. 222–237; read in NOTE-1526.
What was actually shown
The studies: 73 experiments (58 RCTs) with 30,088 participants, searched to November 2018. Each compared a more with a less SDT-supportive arm; effects are Hedges’ g, random effects (pp. 219–222).
End of intervention (Table 1): need support 0.64 [0.35, 0.93], k = 21; autonomy 0.37 [0.15, 0.60], k = 17; competence 0.31 [0.12, 0.49], k = 22; relatedness 0.20 [−0.04, 0.45], k = 14; autonomous motivation 0.30 [0.17, 0.42], k = 37; controlled motivation 0.07, amotivation −0.07 (both ns); health behaviour 0.45 [0.33, 0.57], k = 49; physical health 0.04 (ns); psychological health 0.29 [0.14, 0.45], k = 22.
Follow-up: behaviour 0.28 [0.17, 0.38], k = 28; physical health 0.28 [0.03, 0.53], k = 14; psychological health 0.14 [−0.09, 0.36], k = 10.
Each of these could have been null; the proximal constructs and behaviour were not, and relatedness and the controlled side were.
What the design cannot show. The link between constructs and outcomes is a meta-regression of one study-level effect size on another (Tables 7–8), significant for need support and autonomous motivation on behaviour only after one outlying study is removed. No within-study mediation was tested (p. 237). No study’s outcome was need frustration.
What this does not say
- That the benefits come through need satisfaction. Need satisfaction effect sizes did not predict behaviour effect sizes (Table 7).
- That SDT interventions treat psychopathology. “Psychological health” is a pooled category (quality of life, depression and others) in mostly non-clinical, physical-activity trials, and its effect did not last.
- That more need support gives more benefit. The number of SDT techniques moderated nothing (p. 236).
- That the estimates are unbiased. Blinding was rare, allocation concealment shrank effects, and Egger’s test flags small-study bias for behaviour and psychological health.