A national experiment reveals where a growth mindset improves achievement

12,490 ninth-graders individually randomised across 65 US public high schools drawn as a stratified random sample from a frame of about 12,000 schools

United States; students aged 14-15 entering ninth grade.

The best-designed study the growth-mindset programme has produced, and it is important to say so before deflating it. Individual random assignment within school, an ACTIVE control (a matched brain-functions module of the same length that says nothing about beliefs about intelligence), two 25-minute online sessions about 20 days apart, a preregistered analysis plan, a Tipton generalizability index of 0.98, and unusual independence machinery — ICF drew the sample and obtained administrative records blind to condition, MDRC processed grades blind to treatment implications, and a blinded Bayesian causal-forest reanalysis corroborated the result. The limitation that matters is the outcome - the primary measure is core-course GPA from school administrative records, which is teacher-assigned marks, and there is NO independent standardised test anywhere in the study.

For the preregistered lower-achieving target group (n = 6,320) the effect on ninth-grade core GPA was B = 0.10 grade points (95% CI 0.04 to 0.16, SE 0.03, t = 3.51, p = 0.001), standardised d = 0.11. For higher-achieving students (n = 6,170) it was B = 0.01 (95% CI -0.03 to 0.06, t = 0.48, p = 0.634, d = 0.01) - a clean null. Interaction across the full sample B = 0.09 (0.01 to 0.17, p = 0.034). Rate of D and F grades fell 5.3 percentage points (CI -1.7 to -9.0, p = 0.005). Enrolment in advanced maths in tenth grade rose 3 percentage points (p = 0.001), and that effect applied to all students. Mindset beliefs shifted d = 0.33. School moderation was significant - medium-achieving schools showed larger effects than high-achieving ones - and there was significant cross-school variability in the achievement effect but NOT in the belief-change effect, meaning the intervention changed beliefs everywhere and changed grades only somewhere.

Low. Individual randomisation within school.

Mixed, and the pattern is systematic. The behavioural and course-taking outcomes replicated in Norway by the same author group (Rege et al. 2021). The GPA effect did NOT replicate where the outcome was an independent standardised test - Foliano et al. 2019 in England found exactly zero on national assessments including for disadvantaged pupils, and Ganimian 2020 in Argentina found a precisely estimated zero in the low-achieving population where this study says the effect should be. Macnamara and Burgoyne further note that the published Nature version analyses about half the randomised sample, with 6,222 participants present in the 2018 preprint absent from the published analysis.