The Growth-Mindset Revolution Was Built on 63 Studies. A 98,000-Student Meta-Analysis Found the Average Grade Gain Rounds to Zero.

The Story Everyone Knows

It began with a seductive promise, that telling students their brains grow stronger with effort would lift their grades, and Carol Dweck’s growth-mindset research rode that promise to become the most influential idea in education in decades, traveling from laboratory to classroom with barely a skeptical question asked. MindsetWorks sold “Brainology,” a program of brain animations and testimonials, to school districts across the country, and an Education Week Research Center survey found 53 percent of teachers said using growth mindset was a very important factor in academic achievement, with 85 percent wanting more professional development on the idea. Skepticism barely got a seat.

The pitch worked because the intervention was cheap: a thirty-minute module, a reading, a video, a computer program, could supposedly do what years of patient instruction could not, and districts bought the story that a brief encounter with neuroscience would rewire a student’s relationship with failure while for years almost nobody checked whether any of it worked. In 2023, two psychologists checked properly.

The Study

Brooke Macnamara of Case Western Reserve and Alexander Burgoyne of Georgia Tech assembled the largest evidence audit the field has ever seen, pre-registering the plan at osf.io/ga9jk, publishing their data, and declaring no conflicts of interest and no funding up front. Then they gathered every randomized or quasi-experimental test of growth-mindset interventions on grades and test scores they could find: sixty-three studies, 97,672 students, kindergarten through college.

The headline number was d̄ = 0.05, with a 95% confidence interval of [0.02, 0.09], positive on paper but fragile in practice, and once the authors corrected for publication bias, the documented habit of null results disappearing from journals, the effect dissolved into nonsignificance. Two stricter cuts followed: studies verifying the intervention actually shifted students’ mindsets gave d̄ = 0.04 across thirteen studies and 18,355 students, with a confidence interval of [−0.01, 0.10], and the six highest-quality studies, 13,571 students, gave d̄ = 0.02, CI [−0.06, 0.10], p = .666.

Notice the pattern: the better the study, the closer the effect crept to zero. That slope is the fingerprint of a literature running on bias rather than signal, and the audit found the bias out in the open. Ninety-four percent of interventions had confounds, most often teachers who knew which students received the treatment; authors with a financial stake in mindset products were about two and a half times more likely to report positive effects; and one published study quietly dropped thousands of students whose results pointed the wrong way. None of eleven pre-registered moderators explained the variance. The conclusion, in the authors’ own words: apparent effects “are likely attributable to inadequate study design, reporting flaws, and bias.”

The Arithmetic Nobody Did

Effect sizes in standard deviations stay abstract until translated into something a teacher recognizes, so here is the translation nobody published: take d̄ = 0.02, convert it to percentile ranks, and a median student at the 50th percentile moves to roughly the 50.8th percentile, a gain of less than one percentile rank. The pooled 0.05 lifts the same student to about the 52nd percentile, two ranks. These are rounding errors in lab coats.

Now set them against high-dosage tutoring, where a 2020 NBER meta-analysis of 96 randomized experiments by Nickow, Oreopoulos, and Quan found an average gain of d = 0.37. That moves a median student to roughly the 64th percentile, a gain of about fourteen ranks, and once you do the division the comparison turns stark: tutoring’s effect is 7.4 times the headline mindset figure and 18.5 times the highest-quality estimate. A school that swapped one weekly period of mindset modules for one weekly period of targeted tutoring would, on the published evidence, buy roughly an order of magnitude more learning per student-hour. The math is not close.

The Strongest Counterargument

Defenders of the theory, including Dweck, David Yeager, and Greg Walton, say this meta-analysis graded the wrong assignment. Mindset theory never promised that a thirty-minute module raises every student’s grades; it promised that beliefs about malleability matter in supportive contexts, for students who need them most. Averaging across affluent students in well-resourced schools, where no effect was ever predicted, would drown a real signal among struggling students, and an average near zero is exactly what a heterogeneous theory predicts.

The flagship evidence for that reading is Yeager and colleagues’ 2019 Nature study, a pre-registered, nationally representative randomized trial of roughly twelve thousand U.S. ninth graders. The mindset module lifted lower-achieving students’ GPAs by about 0.10 points: modest, real, and properly randomized. Burnette and colleagues’ separate 2023 meta-analysis reached a compatible conclusion with multilevel methods, finding small effects concentrated in at-risk students when implementation fidelity was high. When Tipton and colleagues challenged the heterogeneity methods in a 2023 commentary, Macnamara and Burgoyne answered with a peer-reviewed rebuttal arguing the critics had changed effect sizes, coding, and study inclusion after the fact. The fairest reading: the intervention is probably not nothing; it is a targeted tool, not a revolution.

What We Didn’t Prove

This meta-analysis tested one claim, that brief mindset interventions raise academic achievement, and its verdict does not travel far beyond it: it did not test whether holding a growth mindset correlates with success, a different and better-supported question. It did not measure motivation, persistence, or mental health, where the competing meta-analysis found larger effects. It cannot rule out benefits for specific students in specific contexts, because an average is blind to heterogeneity by construction, and its bias corrections, though standard, estimate what the literature might show if null results were published rather than certainties about a perfect literature.

The Bottom Line

A belief most American teachers rate as very important to achievement, packaged as software and sold to districts, rests on sixty-three studies whose average causal effect on grades rounds to zero once study quality and publication bias enter the picture. That is no reason to sneer at the idea that effort matters; it is a reason to stop mistaking a good story for strong evidence, and to ask who profits while the confusion lasts.

What You Can Do

If you run a school, audit the hours: a thirty-minute module is cheap, but a weekly advisory period is expensive, so before renewing a vendor contract price the intervention in student-hours and ask what those same hours would buy in tutoring, which the published evidence says is seven to eighteen times more effective per unit of effect.

If you teach, keep the idea and skip the product. Telling students that ability grows with effort costs nothing, while the paid programs added no measurable grade gains on top of the idea, and their vendors were the authors most likely to publish positive results.

If your child is struggling, follow the evidence gradient: the benefits that survived scrutiny concentrated in at-risk students, and the largest documented gains came from tutoring rather than mindset modules. Direct instructional support is the intervention with the evidence behind it.

If you buy education products, demand the three practices whose absence inflated this literature: pre-registration, a check that the intervention changed the thing it claims to change, and published results for the full randomized group. The studies that followed those rules are the ones that found nothing.