Do growth mindset interventions impact students' academic achievement? A systematic review and meta-analysis with recommendations for best practices.

63 studies, N = 97,672 (all studies); 13 studies, N = 18,355 (manipulation check succeeded); 6 studies, N = 13,571 (highest-quality subset)

All ages, international; academic achievement outcomes (GPA, course grades, exam scores).

Preregistered with 11 preregistered moderators and an explicit best-practices quality screen — random assignment, active control, adequate power, a manipulation check, no confounds, no conflict of interest. Where a published version dropped participants relative to an earlier preprint the larger version was used, notably Yeager et al. 2018 rather than the 2019 Nature version, from which 6,222 randomised participants are absent. This is the study that makes the growth-mindset literature auditable.

The quality ladder descends monotonically to zero. All studies d = 0.05 (95% CI 0.02 to 0.09, p = .004). Manipulation check succeeded d = 0.04 (-0.01 to 0.10), not significant. Highest-quality subset d = 0.02 (-0.06 to 0.10), not significant. Egger regression significant with a positive intercept and trim-and-fill imputes 10 missing small or negative studies, after which the overall effect is not significant. NO theoretically meaningful moderator was significant - developmental stage Q(2) = 0.47, p = .791; academic challenge Q(2) = 0.28, p = .870; socioeconomic status Q(1) = 2.10, p = .147; time-to-outcome Q(1) = 0.21, p = .647. The conflict-of-interest analysis is the most quotable result in the tranche - 30% of documents had at least one author with a financial incentive, those articles were 2.5 times as likely to report a significant positive effect (56% vs 21%, chi-square = 7.09, p = .008), and among financially incentivised authors PUBLISHED effects averaged d = 0.18 (0.07 to 0.28) against their own UNPUBLISHED effects of d = 0.02 (-0.02 to 0.06), Q(1) = 7.90, p = .005, a gap absent among non-incentivised authors. Only two studies in the entire literature were both high quality and conflict-free - Foliano et al. 2019 and Polley 2018. And 94% of growth-mindset interventions contained confounds.

Low as a synthesis of randomised trials. The relevant confound here is financial and publication bias, which the authors quantify directly.

Reproduces and strengthens Sisk et al. 2018 on a larger and quality-screened sample. Contested by Tipton et al. 2023 on heterogeneity-modelling grounds and defended in the authors' reply; the disagreement is analytic, not evidentiary, since both sides work from the same studies.