Growth Mindset After the Tests
The claim that one belief permeates every part of a life was printed before any evidence arrived, and it has never been tested. The claim that was tested is small, conditional and cheap, and this is what the national experiment, the two meta-analyses and the English trials actually found.
Answer to the question
Does a growth mindset actually improve grades?
A little, for some students. In the best-designed test, a free online lesson raised core-course grades among the preregistered lower-achieving half by about a tenth of a grade point, roughly double that in low and medium-achieving schools whose pupils already took on challenge, and near nothing on grades in the highest-achieving quarter of schools.
Key takeaways
- In a preregistered experiment on 12,490 American ninth-graders in 65 schools, a free online lesson raised core-course grades among the targeted lower-achieving half by about a tenth of a grade point.
- The effect was concentrated: roughly double that in low and medium-achieving schools with supportive peer norms, and near nothing on grades in the top quarter of schools.
- The two meta-analytic camps diverge on method, over whether to average a highly varied literature or to model its variation, and over whether a preregistered focal subgroup is a finding or a selection.
- The largest English trial, 5,018 pupils in 101 primary schools with trained teachers, found no additional progress on national tests, and the leading proponents concede that teacher-delivered versions have not worked.
Run the cheap online version of a growth mindset lesson, give it to struggling fourteen-year-olds, and expect a small gain. That is what the large tests support. The wider claim has never been tested.
What to do before anything else
The version with evidence behind it is narrow and specific. Two self-administered online sessions of about twenty-five minutes each, three weeks apart, delivered straight to students, with no teacher training. Give it to students at or below their school's median in the term before. Expect core-course grades to move about a tenth of a grade point, and the share of those students finishing the year below a 2.0 average to fall by just over five percentage points.
The limit travels with the number. Those figures are the national experiment's preregistered result for its lower-achieving half, 6,320 students out of the 12,490 it randomised and analysed. The sceptical camp counts that focus as leaving the other half out of the headline. The authors answer that they named the target group in advance and reported it as planned.
The version without evidence is the expensive one. In the Education Endowment Foundation's first English trial, the teacher-training arm cost sixteen pounds per pupil a year and moved nothing. The pupil-workshop arm cost three hundred and ninety-seven pounds per pupil a year and produced a two-month gain the Foundation itself reports as not statistically significant.
The cheap arm did nothing and the dear arm did something unreliable.
The cheap arm did nothing and the dear arm did something unreliable.
Carolina Gazmuri, whose 2025 review rates each trial by sample size, missing data and data quality and excludes the national study for reporting only a subgroup, closes on the practical point anyway:
> "if there is an opportunity to implement such interventions at a minimal or negligible cost, or as part of another objective, it might be reasonable to proceed with them, considering the potential for a small positive impact."
That is the ceiling of the practical case, and the rest of this article is why it sits so low.
Where the idea reached the public
Mindset, published in 2006 and revised in 2017, is where growth mindset entered general circulation. It is a primary source for what Carol Dweck claims, and it is not where this article's evidence comes from, because it reports no effect size and no sample size for any study of its own.
The paper behind the founding praise experiment holds what the book leaves out, and its first author is Claudia Mueller. Six studies, 412 fifth-graders, the largest sample 128. The misrepresentation finding the book rounds to almost forty per cent is thirty-eight per cent of ability-praised children against thirteen per cent of the comparison group, in a study of eighty-eight. [What Forty-One Participants Buys](https://thewhyweproject.com/articles/what-forty-one-participants-buys) sets out what samples that size can settle.
The one direct independent replication we could locate is Yue Li and Timothy Bates, three studies across 624 children in Harbin. The praise effect fell short of conventional significance in the first, was absent in the second, and in the third produced a gain that an active control beat by about a third. That control told children ability cannot be changed while recognising their work.
Li and Bates designed each study to have an eighty-five per cent chance of detecting a moderate effect. David Yeager and Carol Dweck answer that each was still underpowered on its own, that the straightforward treatment of underpowered replications is to pool them, and that pooled across the three the original result replicates at conventional significance. Both readings are on the record.
What the national study found
The best test this idea has had is the National Study of Learning Mindsets, first-authored by David Yeager and published in Nature in 2019. Twelve thousand four hundred and ninety ninth-graders in 65 American public high schools, individually randomised inside their schools. Generalisability was measured, at an index of 0.98 against eight benchmarks, where 0.90 corresponds to essentially a random sample.
The study also bought its own independence. One research company collected the data blind to condition and a second processed it blind, keeping the right to publish a dissenting report. The analysis plan was preregistered with exactly two moderation hypotheses, and a separate team of statisticians reanalysed a blinded dataset.
In the preregistered lower-achieving half, 6,320 of those students, core-course grades rose by a tenth of a grade point, around a tenth of a standard deviation. The share of them finishing below a 2.0 average fell by just over five percentage points, from forty-six per cent to just under forty-one.
Where it landed matters more than that it landed. Effects were smaller in higher-achieving schools, and the second preregistered moderator was behavioural: challenge-seeking norms estimated from what control-group students did on a make-your-own-worksheet task.
| Who | What moved | By how much |
|---|---|---|
| Lower achievers, all 65 schools | Core-course grades | A tenth of a grade point |
| Lower achievers, low and medium schools, strongest norms | Core-course grades | Close to a fifth of a grade point |
| Lower achievers, top quarter of schools | Core-course grades | Near nothing |
| Full cohort, the 41 schools with course records, exploratory | Advanced maths enrolment | Three percentage points, from thirty-three to thirty-six |
| Top quarter of schools, same 41, exploratory | Advanced maths enrolment | Four percentage points, against two in the rest |
The enrolment result runs the opposite way to the grades result, and the authors gloss it themselves: an exclusive focus on grade averages would have obscured the benefits to students in higher-achieving schools. Our reading is the one the two rows force, that this was not a single effect at all. The analysis was not preregistered and rests on the 41 schools with usable course records. The authors add their own brake: not all forms of growth mindset intervention can be expected to raise grades, even in the targeted groups.
What the two meta-analytic camps measured
Brooke Macnamara and Alexander Burgoyne's 2023 review in Psychological Bulletin averages the literature study by study. Across 63 studies and 97,672 students the average effect is a twentieth of a standard deviation, which stops being distinguishable from zero once publication bias is corrected for. Across the thirteen studies that demonstrably shifted mindsets it is a twenty-fifth, and across the six strongest a fiftieth. Their sharpest point is about design: on their coding, 94 per cent of these interventions differ from their controls in something besides the mindset message.
Jeni Burnette and colleagues, in the same journal and issue and read here at its abstract, covered 53 independent samples and modelled the variation in effects where the other camp averaged it. In subsamples expected to benefit most, with high implementation fidelity, academic achievement moved by about one seventh of a standard deviation. They printed the prediction interval beside it, running from a little below zero to about a third of a standard deviation, and wrote that extensive variation should be expected from future interventions.
On the positive camp's own estimate, the next school could get nothing.
Elizabeth Tipton's commentary, abstract only, names the fork: aggregate effect sizes study by study and split moderators one at a time, or model the variation within studies. She reports that applying the second approach to the first camp's data recovers a meaningful effect in at-risk groups, and labels that reanalysis exploratory herself. Macnamara and Burgoyne reply that it altered coding and inclusion.
The second fork is which version of the national study to analyse, and it is a dispute about data as much as method. Macnamara and Burgoyne used the 2018 preprint rather than the 2019 paper, because 6,222 participants were missing from the published version. Those 6,222 are the other half of the same 12,490 and not a further group, which is the arithmetic the dispute turns on. Gazmuri excluded the study on the same ground. Against that, the half left out is the preregistered non-target group, named before the data were seen, and Burnette's team made focal subsamples an explicit moderator.
What the British trials returned
The hardest number in this literature is English. Five thousand and eighteen pupils in 101 primary schools, an eight-week programme of up to two and a half hours a week, with headteachers and Year 6 teachers trained to deliver and embed it. The Education Endowment Foundation's summary is that the evaluation found no evidence that participating pupils made any additional progress in literacy or numeracy on national Key Stage 2 tests.
The effect sizes Gazmuri reads off that report, which we take from her read and not our own, are minus one hundredth of a standard deviation in maths and zero in reading and in grammar, punctuation and spelling, with the Foundation's wording taken from its own press release. One caveat cuts against the null: many control-school teachers already knew the theory, so a contaminated comparison makes this weaker evidence of nothing than it looks.
Both camps agree about the British record.
Yeager and Dweck wrote in 2020 that teacher-focused growth mindset interventions have not worked, naming two English trials developed with great care and labour-intensive to run.
Their best argument is about the yardstick, and we find it persuasive. Through their account of Kraft's 2020 work, every study funded by the United States federal Investing in Innovation programme had to pre-specify outcomes and report them whatever they showed. The median effect was three hundredths of a standard deviation and the ninetieth percentile twenty-three hundredths, with nothing above that among adolescents.
Beside it they set, from their own calculations, energy-use norms at three hundredths, flu-vaccination plans at twelve hundredths and a simplified financial aid form at sixteen hundredths. A free lesson landing inside that range is doing ordinary work.
So we take the heterogeneity reading, at the size it was found, and add three checks for anyone being sold this.
First, ask which claim is on offer. A scripted online lesson that nudges the grades of struggling teenagers is a different proposition from a belief that permeates every part of a life, and only the first has been put to a test. It is also why this is not the replication story that [The Psychology You Learned Didn't Replicate](https://thewhyweproject.com/articles/the-psychology-you-learned-didnt-replicate) tells: the finding shrank and specialised under better methods, and it did not evaporate.
Second, ask who was in the sample and who moved. A growth mindset result quoted without an age group, an achievement band and a school type has dropped the part that mattered most in the best study anybody has run. The same programme's reach into adult self-control, which [Willpower Is Not a Fuel Tank](https://thewhyweproject.com/articles/ego-depletion-debunked) covers, is a separate question.
Third is price and what it displaces, which is where the whole practical case lives. A free online module given to struggling fourteen-year-olds is worth the hour it takes at the small and conditional size the national experiment found, while the several hundred pounds per pupil that a staff-trained version costs buys a result two English trials have already measured, and what they measured was nothing at all on the national tests.
Common questions
Does a growth mindset actually improve grades?
A little, for some students, in some schools. In the national experiment, core-course grades in the preregistered lower-achieving half rose about a tenth of a grade point. That roughly doubled where school achievement was low or medium and pupils already treated difficulty as normal, and came to near nothing in the top quarter.
Did the growth mindset research fail to replicate?
Not in the way several famous findings did. The founding praise studies are small and the one direct independent replication came out mixed. But the field then ran a large preregistered national experiment, and the effect got smaller and more conditional. The camps also dispute which version of that experiment to analyse: the sceptics used its 2018 preprint because participants were missing from the published paper.
Should schools train teachers in growth mindset?
Not as a standalone way of raising attainment, on current evidence. The largest English trial trained teachers across 101 primary schools and found no additional progress on national tests, and the leading proponents concede in print that teacher-focused versions have not worked. The version with results behind it goes straight to students.
Sources
Yeager, D. S., Hanselman, P., Walton, G. M., Murray, J. S., Crosnoe, R., Muller, C., Tipton, E., Schneider, B., Hulleman, C.
Mueller, C. M., & Dweck, C. S. (1998). 'Praise for intelligence can undermine children's motivation and performance.' Journal of Personality and Social Psychology, 75(1), 33-52.
Li, Y., & Bates, T. C. 'Does mindset affect children's ability, school achievement, or response to challenge? Three failures to replicate.' [FULL MANUSCRIPT READ AT SOURCE, the copy at mrbartonmaths.com; data and code at osf.io/uethh.
Macnamara, B. N., & Burgoyne, A. P. (2023). 'Do Growth Mindset Interventions Impact Students' Academic Achievement? A Systematic Review and Meta-Analysis With Recommendations for Best Practices.' Psychological Bulletin, 149(3-4), 133-173.
Macnamara, B. N., & Burgoyne, A. P. (2023). 'A Spotlight on Bias in the Growth Mindset Intervention Literature: A Reply to Commentaries.' Psychological Bulletin, 149(3-4), 242-258.
Sisk, V. F., Burgoyne, A. P., Sun, J., Butler, J. L., & Macnamara, B. N. (2018). 'To What Extent and Under Which Circumstances Are Growth Mind-Sets Important to Academic Achievement? Two Meta-Analyses.' Psychological Science, 29(4), 549-571.
Burnette, J. L., Billingsley, J., Banks, G. C., Knouse, L. E., Hoyt, C. L., Pollack, J. M., & Simon, S.
Tipton, E., Bryan, C., Murray, J., McDaniel, M. A., Schneider, B., & Yeager, D. S. (2023).
Yeager, D. S., & Dweck, C. S. (2020). 'What can be learned from growth mindset controversies?' American Psychologist.
Education Endowment Foundation (2019). 'Changing Mindsets: Effectiveness trial', independently evaluated by the National Institute of Economic and Social Research; Foliano and colleagues.
Education Endowment Foundation (2014/2017). 'Changing Mindsets, first trial', University of Portsmouth, independently evaluated by NIESR; published as Rienzo and colleagues 2015.
Gazmuri, C. (2025). 'Can growth mindset interventions improve academic achievement? A structured review of the existing evidence.' Review of Education, 13, e70066.
Dweck, C. S. (2017). Mindset: Changing the Way You Think to Fulfil Your Potential, revised edition.