---
title: Why Two Growth Mindset Meta-Analyses Disagree - Linus Murphy
description: Two 2023 Psychological Bulletin meta-analyses reached different
  verdicts on growth mindset. The gap comes from what they counted and how they
  counted it.
url: https://linusmurphy.com/news/growth-mindset-meta-analyses-disagree/
language: en-GB
type: article
author: Linus Murphy
author_url: https://linusmurphy.com/about/
published: 2026-10-22
modified: 2026-10-22
---

[News](https://linusmurphy.com/news/)

# Two growth mindset meta-analyses disagree. Here is why

By [Linus Murphy](https://linusmurphy.com/about/) • 22 October 2026

![Two compasses on a weathered map, their needles pointing in slightly different directions](https://linusmurphy.com/images/news/growth-mindset-meta-analyses-disagree.webp)

In 2023 *Psychological Bulletin* published two meta-analyses of growth mindset interventions that reached different verdicts. One concluded the apparent effects were likely due to weak studies and bias. The other found modest benefits for targeted groups. The gap came from what each counted and how. For leaders, it is a lesson in reading any evidence before acting on it.

The sceptical review is [Brooke Macnamara and Alexander Burgoyne’s meta-analysis of growth mindset interventions and students’ academic achievement](https://pubmed.ncbi.nlm.nih.gov/36326645/). I read its abstract and the authors’ full preprint on PsyArXiv. The other is [Jeni Burnette and colleagues’ review](https://pubmed.ncbi.nlm.nih.gov/36227318/), which I cover in depth in [my piece on for whom growth mindset interventions work, and why](https://linusmurphy.com/news/growth-mindset-interventions-for-whom-and-why/). Its full text is paywalled, so its figures here come from the abstract and from how others in the exchange describe it.

## What did Macnamara and Burgoyne find?

They found a small overall effect on academic achievement that disappeared after correcting for possible publication bias, and no significant effect in the highest-quality studies. They concluded that “apparent effects of growth mindset interventions on academic achievement are likely attributable to inadequate study design, reporting flaws, and bias.”

Their question was twofold. Do growth mindset interventions generally improve students’ achievement? And if they do, is a changed mindset the reason, or are the effects produced by how studies were designed, reported and published?

They ran three meta-analyses, each with a stricter filter.

| Analysis | Studies | Students | Effect (d) |
| --- | --- | --- | --- |
| All studies meeting the inclusion criteria | 63 | 97,672 | 0.05 (95% CI 0.02 to 0.09), not significant after bias correction |
| Studies showing the intervention changed mindsets | 13 | 18,355 | 0.04 (95% CI −0.01 to 0.10) |
| Highest-quality evidence | 6 | 13,571 | 0.02 (95% CI −0.06 to 0.10) |

They also reported, in the abstract, that “authors with a financial incentive to report positive findings published significantly larger effects than authors without this incentive.” In the preprint, articles with at least one such author were more than two and a half times as likely to report a significant positive result: 56% against 21%.

Their review built on an earlier one. In 2018 Victoria Sisk, with Burgoyne and Macnamara among the co-authors, [meta-analysed 43 intervention studies with 57,155 people and found weak overall effects](https://pubmed.ncbi.nlm.nih.gov/29505339/), with some support for benefits among students with low socioeconomic status or who were academically at risk.

## What did Burnette and colleagues find instead?

They found small, highly variable effects that were larger when interventions were well delivered to the groups expected to benefit. For those groups the abstract reports d = 0.14 for academic achievement and d = 0.32 for mental health.

Their review covered randomised trials from 2002 to the end of 2020, across 53 independent samples. It looked at academic achievement, mental health and social functioning. It reported a 95% prediction interval for achievement of −0.08 to 0.35: the range of effects to expect from the next intervention.

They also framed the task differently. As quoted by Tipton and colleagues, Burnette’s team argued that “heterogeneity in effects is expected for growth mindset interventions” and that meta-analyses in such fields “should not try to deliver” a “simple verdict.” Their stated aim was “clarifying for whom these interventions work best.” Macnamara and Burgoyne, by contrast, asked first whether the interventions work in general.

Read side by side, the headline numbers are not wildly apart. Both are small. The difference is in what each team thought the numbers meant.

## Which studies did each review count?

Different ones, by design. Macnamara and Burgoyne looked only at students and academic achievement, and accepted any comparison group. Burnette and colleagues required randomisation and looked at several outcomes, not just grades.

The Macnamara and Burgoyne preprint sets out its rules. A study qualified if a growth mindset treatment, not combined with any other intervention, was given directly to students; if there was a comparison group, whether active, passive or taught a fixed mindset; and if it reported a grade, exam or standardised test result. They excluded studies that only praised effort without teaching that ability can change. Their search ran to 7 August 2019 and found 63 studies with 79 independent samples.

Burnette and colleagues covered articles published up to the end of 2020, but required a randomised design and allowed outcomes beyond achievement.

In their later reply, Macnamara and Burgoyne say the two datasets are a main difference. By their count, [Burnette and colleagues included 32 studies of academic achievement, about half as many as their own 63](https://osf.io/preprints/psyarxiv/embhs).

## How did each review treat studies with more than one result?

This is the technical heart of the disagreement. Macnamara and Burgoyne combined the results within each study into a single effect. Burnette and colleagues kept the separate results and modelled the variation between groups inside each study.

Many mindset studies report several results: for struggling and strong students, for different schools, for different tests. Combining them gives one tidy number per study. Keeping them separate lets you ask which group moved.

Elizabeth Tipton, Christopher Bryan, Jared Murray, Mark McDaniel, Barbara Schneider and David Yeager [argued in a commentary in the same issue](https://pmc.ncbi.nlm.nih.gov/articles/PMC10495100/) that this choice drove much of the gap. They note that Macnamara and Burgoyne used software that handles one effect per study, while Burnette and colleagues used multilevel models that keep every relevant result.

Their sharpest example is a large World Bank trial in Peru. By their account, it reported effects of 0.23 to 0.35 standard deviations in high-poverty schools and roughly zero in low-poverty schools. Averaged into a single figure, the study looks like it did nothing.

The national US study shows the same issue. David Yeager and colleagues reported in *Nature* in 2019 that [a short online intervention improved grades among lower-achieving students](https://pubmed.ncbi.nlm.nih.gov/31391586/). Tipton and colleagues point out it was designed to find effects only in that half of the sample.

## How did each review judge quality and bias?

Macnamara and Burgoyne coded every study against a preregistered list of best practices and used standard publication-bias tests. Burnette and colleagues treated quality as a continuous control and used a different family of bias models. Each side criticises the other’s choices.

Macnamara and Burgoyne’s criteria included an active control group, no other differences between conditions, a power analysis planned in advance, random assignment of individual students, blinding, a before-and-after check that mindsets actually changed, preregistration, complete reporting and no author with a financial interest. No study met all of them. In the preprint, 70% of samples met half or fewer. The highest-quality analysis rested on six studies that had shown a change in mindsets and met at least 60% of the criteria.

Tipton and colleagues object to several of these definitions. They argue that requiring individual random assignment penalises school-level trials, which are common and appropriate in education. They also object that the financial-interest coding counted income an author earned after a study was published.

On bias, Tipton and colleagues argue that the funnel-based tests Macnamara and Burgoyne used can signal publication bias that is not there when effects genuinely vary. They report that Burnette and colleagues used selection models and found no evidence of publication bias.

Macnamara and Burgoyne answer that their hypotheses, moderators and methods were preregistered, and Burnette and colleagues’ were not. They quote Burnette and colleagues: “we did not preregister the search protocol or decisions outlined below.”

## What did the published reanalysis and reply add?

They sharpened the disagreement rather than settling it. Tipton and colleagues reran Macnamara and Burgoyne’s data with Burnette’s methods and found an effect for at-risk groups. Macnamara and Burgoyne replied that the reanalysis changed the data as well as the method.

Tipton and colleagues used all 122 effect sizes rather than one per study. They reported a mean effect of 0.09 standard deviations, an effect of 0.15 among at-risk groups and a prediction interval of −0.22 to 0.40, close to Burnette’s. They say several of them worked on the national US study, and they posted their data and code.

Macnamara and Burgoyne’s reply, which I read as the accepted manuscript on PsyArXiv, says Tipton and colleagues altered information beyond separating the results in 22 of the 63 studies. They list changed effect sizes, changed risk codings and studies regrouped, often without explanation.

They also question the focal-group approach. In Burnette’s review, they write, focal groups were the ones the original study authors had identified, and these mixed very different characteristics from study to study. Subgroups chosen that way can be picked after the fact.

The reply also raises interests. It notes that Tipton and several co-authors are part of an initiative to implement growth mindset interventions in classrooms, and that Yeager is registered with a speakers’ bureau.

## What do the two sides agree on?

More than the headlines suggest. Both report small average effects and wide variation between studies, and both coded study quality rather than taking the literature at face value. They disagree about whether larger effects for targeted groups are a genuine pattern or an artefact of how those groups were chosen.

Macnamara and Burgoyne’s reply quotes Burnette and colleagues acknowledging that “null and even negative (in the case of academic achievement) effects are to be expected in growth mindset interventions.” No one in the exchange is claiming a large, universal effect.

## What is Linus Murphy’s take?

My take is that this exchange is a better lesson in judgement than in mindset. Two careful teams looked at overlapping evidence and reached different conclusions because they made different, defensible choices about what to count and how. Leaders face that problem every week with their own data.

It sits in [the “Know yourself first” cluster of my Executive Coaching Methodology](https://linusmurphy.com/guide/#know-yourself-first), because the risk is internal: we read evidence through what we already believe. Each side in this debate accuses the other of exactly that, and each has a point worth hearing.

The closest part of the 30-part series is [Self and team awareness](https://linusmurphy.com/guide/leadership-from-the-inside-out/self-and-team-awareness/), where I argue awareness is a measurement habit, not a trait. The meta-analyses show why. How you measure decides what you find. In my [7-step coaching process](https://linusmurphy.com/services/impactful-leadership-7-step-executive-coaching-process/), step two, Awareness, exists to replace guesswork with evidence, and that only works if you are honest about how the evidence was gathered.

[Winning is an inside job](https://linusmurphy.com/guide/leadership-from-the-inside-out/winning-is-an-inside-job/) adds the discipline of preparation. Decide in advance what result would change your mind. Macnamara and Burgoyne’s case for preregistration is the research version of that habit.

For teams, I come back to [Attitude: what difference does it make?](https://linusmurphy.com/guide/building-winning-teams/attitude-what-difference-does-it-make/), where I first set these two reviews side by side. My conclusion there stands: anyone selling a mindset workshop as a performance lever for a senior team is ahead of the data.

When a board meets two credible analyses that disagree, I suggest four questions. What was counted, and what was left out? Were several results inside one study averaged or kept apart? How was quality judged, and was that decided before the data came in? Who has a stake in each answer?

The same questions work on an internal evaluation of a leadership programme. A pattern I see often is a single average reported to the board, with no one asking which people actually changed. That is exactly the argument at the centre of this exchange.

If your executive team needs a sharper way to argue about evidence without turning it into a contest of loyalties, that is part of my [executive team coaching](https://linusmurphy.com/services/executive-team-coaching/) work. It also links to how a team builds [psychological safety at the executive table](https://linusmurphy.com/articles/psychological-safety-at-the-executive-table/), so people can challenge a favoured number without paying for it.

## Questions I am asked

**What did Macnamara and Burgoyne conclude about growth mindset interventions?**

They concluded that apparent effects on students' academic achievement are likely due to inadequate study design, reporting flaws and bias. Across 63 studies and 97,672 students the overall effect was small, d = 0.05, and not significant after correcting for possible publication bias.

**Why did the two 2023 meta-analyses reach different conclusions?**

Mainly because they counted different studies and analysed them differently. Macnamara and Burgoyne combined each study's results into one effect and judged it against strict quality criteria; Burnette and colleagues kept the separate results within studies and asked which groups benefited.

**Did the reanalysis by Tipton and colleagues settle the debate?**

No. Tipton and colleagues reported that applying Burnette's methods to Macnamara and Burgoyne's data produced a significant effect for at-risk groups. Macnamara and Burgoyne replied that the reanalysis also changed effect sizes, risk codings and study groupings in 22 of their 63 studies, often without explanation.

**What do both sides agree on?**

That average effects are small and that results vary a great deal from study to study. Both reviews also coded study quality rather than taking the literature at face value. They disagree about whether the larger effects in targeted groups are real.

**What should a board take from a research dispute like this?**

Ask what was counted, how multiple results were combined, how quality was judged and who has a stake in the answer. The same four questions apply to any internal evaluation of a development programme.

## Sources

1.  [Do growth mindset interventions impact students' academic achievement? A systematic review and meta-analysis with recommendations for best practices](https://pubmed.ncbi.nlm.nih.gov/36326645/) — Psychological Bulletin (PubMed), 3 November 2022
2.  [A systematic review and meta-analysis of growth mindset interventions: For whom, how, and why might such interventions work?](https://pubmed.ncbi.nlm.nih.gov/36227318/) — Psychological Bulletin (PubMed), 13 October 2022
3.  [Why meta-analyses of growth mindset and other interventions should follow best practices for examining heterogeneity: Commentary on Macnamara and Burgoyne (2023) and Burnette et al. (2023)](https://pmc.ncbi.nlm.nih.gov/articles/PMC10495100/) — Psychological Bulletin (PubMed Central), March 2023
4.  [A spotlight on bias in the growth mindset intervention literature: A reply to commentaries that contextualize the discussion (Oyserman, 2023; Yan & Schuetze, 2023) and illustrate the conclusion (Tipton et al., 2023)](https://osf.io/preprints/psyarxiv/embhs) — PsyArXiv (accepted manuscript for Psychological Bulletin), 23 July 2023
5.  [To What Extent and Under Which Circumstances Are Growth Mind-Sets Important to Academic Achievement? Two Meta-Analyses](https://pubmed.ncbi.nlm.nih.gov/29505339/) — Psychological Science (PubMed), 5 March 2018
6.  [A national experiment reveals where a growth mindset improves achievement](https://pubmed.ncbi.nlm.nih.gov/31391586/) — Nature (PubMed), 7 August 2019

![Portrait of Linus Murphy](https://linusmurphy.com/images/portraits/linus-headshot.jpg)

**[Linus Murphy](https://linusmurphy.com/about/)**

Elite executive coach