Growth mindset: what the big trials found.

Few research programs show the full life cycle of a psychological idea as completely as this one. An elegant lab insight; a decade of promising small studies; a popular wave that outran the data; a scale-up era of preregistered trials; a meta-analytic reckoning; and a methods fight that clarified what everyone had actually shown. Whatever one concludes about mindset itself, the arc is the field’s best teaching case for a question every training buyer eventually faces. What happens to an exciting effect when someone finally tests it properly?

The idea is genuinely elegant. People differ in what the field calls implicit theories of ability — unspoken beliefs about what ability is. Some treat intelligence as a fixed quantity to be displayed; others treat it as a capacity that grows with effort. Those theories shape how people respond to difficulty. Fixed-theory learners avoid challenge and fold after failure; growth-theory learners treat setbacks as information (Dweck & Leggett, 1988).

Early studies delivered striking demonstrations. Praising children’s intelligence rather than their effort made them shy of challenge and fragile (Mueller & Dweck, 1998). Teaching teenagers that the brain grows with use appeared to bend grade curves upward (Blackwell, Trzesniewski & Dweck, 2007). Large correlational datasets — beliefs and grades tracked together — seemed to agree, with mindset tracking achievement even across a national cohort (Claro, Paunesku & Dweck, 2016).

By the mid-2010s the idea had outgrown psychology entirely. School districts adopted mindset curricula. Ministries wrote it into national education strategies. The corporate world turned it into workshops and values decks. Adoption at that scale usually ends the scientific story. Here it started the interesting part.

Then the idea did what few psychological ideas ever do: it went to genuinely large, preregistered trials — studies whose methods and analyses were locked in before the data arrived. What came back is one of the most instructive datasets in applied psychology. Less because of what it says about mindset than because of what it says about how ideas should be tested before they become policy.

What the intervention actually is

Before weighing the effects, it helps to know what is actually delivered, because the treatment is far more modest than the movement built around it. A standard mindset intervention is one or two online sessions totaling under an hour. Students read a short article explaining that the brain is malleable — that ability grows with effort, strategies, and help from others — framed with brain-science imagery of strengthening connections. Then comes the quietly clever part, a persuasion technique the literature calls “saying-is-believing.” Rather than being tested on the message, students write a letter explaining it to a struggling younger student. Arguing for the position, the theory goes, installs it more deeply than receiving it (Yeager et al., 2019).

Knowing the dose reframes both camps’ claims. That such a treatment could move anything measurable months later is genuinely remarkable. The targeted national-trial effect deserves its due as one of behavioral science’s better value-for-money stories. But the same modesty explains the nulls. A fifty-minute reading-and-writing exercise is competing against a student’s entire accumulated experience of school — and against every daily signal about whether effort actually pays where they sit. The intervention is a nudge, in the strict sense: nudges move outcomes at the margin, for people near a threshold, when the environment lets the nudged behavior succeed — which is, almost word for word, the moderator pattern the big trials found.

The meta-analyses arrive

The first full audit ran two meta-analyses side by side. The mindset–achievement correlation, across hundreds of thousands of students: r ≈ .10. Real — and a tenth of what the enthusiasm implied. Mindset interventions, across 43 effect sizes: d ≈ 0.08 on average, statistically distinguishable from zero only in spots. Effects were somewhat larger (d ≈ 0.19) for academically high-risk and poorer students (Sisk, Burgoyne, Sun, Butler & Macnamara, 2018). The pattern — tiny average, non-zero pockets — set the terms for everything since.

Independent large-scale replications sharpened the skeptical case. A UK trial across schools, run by the Education Endowment Foundation, delivered the intervention as designed — and found no detectable effect on attainment (Foliano, Rolfe, Buzzeo, Runge & Wilkinson, 2019). Careful replication work in China failed to reproduce the foundational praise and mindset results in that population (Li & Bates, 2019). And when the intervention literature was re-graded under best-practice rules — preregistration, unbiased outcome measures, sound analysis — the estimated average effect fell to about zero (Macnamara & Burgoyne, 2023).

The strongest trial — and the honest reading

The centerpiece of the defense is the National Study of Learning Mindsets. It was a preregistered randomized trial of a one-hour online intervention, run on a nationally representative sample of over 12,000 US ninth-graders. The result: no meaningful effect for students overall. Among lower-achieving students, an average gain of about 0.10 grade points — with the effect concentrated in schools whose peer norms supported challenge-seeking (Yeager et al., 2019). It is at once the best evidence the intervention has and a demolition of the universal version. A free one-hour treatment with a small targeted effect is a good public-health-style tool — and it is not a theory of learning.

The founding camp’s own look-back concedes the reframing. The interesting question was never “does it work on average?” but “for whom, and under what conditions?” On this view, the variation in effects — the heterogeneity — is the central finding rather than an embarrassment (Yeager & Dweck, 2020). The skeptics’ reply: a theory rescued by moderators must predict them in advance, not discover them in the wreckage of null averages (Macnamara & Burgoyne, 2023). Both positions are worth teaching. Between them sits most of what applied psychology has learned about scaling lab effects.

The school-norms moderator deserves a moment of its own, because it carries the deepest lesson in the dataset. The intervention worked best not in the most struggling schools but in schools whose peer culture supported challenge-seeking (Yeager et al., 2019). These were places where a student who believed effort pays could act on the belief without social penalty. And acting on it met teaching that rewarded it.

A belief, in other words, is only as effective as the environment’s willingness to make it true. Persuade someone that ability grows, then drop them into a context where extra effort meets the same stale material and the same fixed judgments. The belief either dies of contradiction or survives as decoration. The moderator is not a footnote to the mindset story; it is the mechanism showing through. And it transfers directly to any workplace tempted to buy the message without renovating the conditions.

The praise result: the piece that survives

While the intervention estimates shrank, one branch of the program held its value. It is worth understanding why. The praise experiments were not belief-change studies at all — they were feedback-design studies. Fifth-graders solved problems and were praised either for the trait (“you must be smart at this”) or for the process (“you must have worked hard”). What followed diverged sharply, across multiple replications within the original paper. Trait-praised children avoided challenge, and their performance fell apart after failure (Mueller & Dweck, 1998).

Notice the structural differences from the intervention literature. The praise is delivered at the moment of performance, by the environment itself, repeatedly — not once, in a module, months before the outcome. It requires no belief to be installed and defended against a contrary environment; it is the environment. And its mechanism runs through the feedback channel, where task-versus-person effects are among the most replicated results in applied psychology. That is why this finding lines up with the independent feedback literature while the standalone modules wobbled: it was never really a mindset result — it is a feedback result wearing the mindset program’s colors. And it transfers with full force to adult workplaces, where person-level praise and person-level criticism remain the default dialect of management.

The corporate import problem

Reading the two camps side by side is also a template for judging every other contested program this library covers — ask four questions of any claimed effect. How big is it on average, in the largest preregistered tests? For whom is it larger — and were those moderators predicted, or dug up afterward? What dose and conditions did the successful trials actually use, and what happens to the estimate as study quality rises? Mindset answers: small; lower-achievers in supportive contexts, partly predicted; an hour of online persuasion plus an environment that rewards it; it shrinks. Any vendor claim that cannot survive the same four questions has told you what you need to know.

Everything above concerns school students — the only population with large trials. The version most professionals actually meet is the corporate import: mindset workshops, culture decks, performance-review vocabulary. That version has essentially no outcome evidence behind it at all.

The one nearby finding worth keeping came from the praise studies. Person-level judgment (“you’re brilliant”) measurably degrades later risk-taking and resilience, compared with process-level feedback (Mueller & Dweck, 1998). The result survives because it is a feedback-design finding, not a belief-change finding — and it lines up with the entire feedback literature. Meanwhile, the meta-analytic reviews of the genre urge exactly the caution vendors skip: effects that vary widely, depend on delivery, and are real mainly for at-risk groups under supportive conditions (Burnette et al., 2023).

Measurement trouble under the whole edifice

One more layer of the controversy deserves daylight, because it sits beneath both camps. The mindset itself is measured by self-report: agreement with a handful of statements like “you can always greatly change how intelligent you are.” Everything in the correlational literature inherits that instrument’s weaknesses. People differ in how they read the items and in their willingness to give the culturally favored answer. And the growth answer has been the favored one in English-speaking schools for a decade — which alone could squeeze the scale’s variance and its correlations.

Causal direction is unresolved in the correlations too. Believing ability grows may raise achievement. But achieving visibly is also an excellent way to acquire the belief. The modest r ≈ .10 is compatible with either arrow, or both (Sisk et al., 2018).

None of this is fatal. The randomized trials exist precisely because the correlations could not carry causal weight. But it disciplines how the surviving numbers should be read. The defensible chain is narrow: a brief persuasive exercise, delivered to teenagers near an achievement margin, in environments that reward the persuaded behavior, moves grades slightly. Every widening of that chain — to adults, to averages, to cultures where the belief landscape differs, to outcomes beyond grades — is a stretch the measured evidence has, so far, declined to fund.

What the evidence doesn’t show

• It doesn’t show mindset is fake. The correlation is real and the targeted national-trial effect is real. What failed was the universal claim — mindset as a major lever for everyone.
• It doesn’t show cheap is bad. An hour-long online module producing d ≈ 0.10 for struggling students is excellent cost-effectiveness. The error is upgrading “cheap and slightly helpful for some” into “transformative for all.”
• It doesn’t validate mindset-washing. No trial supports posters, slogans, or relabeling layoff resilience as a growth opportunity. Belief change without changed practice conditions is precisely the arm that fails (Foliano et al., 2019).
• It doesn’t transfer automatically to adults at work. The trial base is adolescents in schools. Corporate claims are extrapolation, and should be priced as such.

Where the evidence stops

1. 1It doesn’t show mindset is fake
2. 2It doesn’t show cheap is bad
3. 3It doesn’t validate mindset-washing
4. 4It doesn’t transfer automatically to adults at work

© 2026 FUTURE PROOF™

What this means for practice

Spend the mindset budget on the conditions a growth mindset would need to be true. A learner’s belief that they can improve is rational exactly when improvement is real, visible, and rewarded. That makes mindset a practice-design and measurement problem wearing a messaging costume. Build training where effort reliably produces measurable gains: retrieval instead of re-watching, spacing instead of cramming, feedback that carries information, practice aimed at diagnosed weaknesses. In that environment, the growth message becomes redundant in the best way. Learners infer it from their own path — the one messenger no one argues with.

Then close the loop the trials point at: show people their own data. Mastery curves, before-and-after diagnostics, the list of things they could not do last quarter — evidence of growth, delivered to the person who grew, the honest descendant of the “brain is malleable” article. Keep the one import that replicates everywhere: feedback about process and task, never about the person, in performance reviews as much as in courseware. Skip the posters — the norms moderator says environments teach mindsets better than messages do. And when a vendor quotes the early effect sizes at your leadership team, quote Figure 1 back. The average effect is a rounding error; the targeted effect is real and small; the version being sold is usually the one the evidence retired.