# What we learned about the mind

2016–2026 · Six lessons from psychological research over the past decade. Effects depend on context, beliefs do not always predict action, and credible findings need independent checks.

## 1. A published finding still needs an independent test.

A large 2026 replication project reported success estimates from 28.6% to 74.8%, depending on which of 13 criteria was used. Even the definition of a successful replication changes the answer.

Look for independent replications and the size of the effect. These percentages describe the sampled social and behavioural studies, not the proportion of all psychology that is true.

[Investigating the replicability of the social and behavioural sciences](https://www.nature.com/articles/s41586-025-10078-y)

## 2. A growth mindset can help, within a supportive setting.

A US school experiment found that a brief intervention about developing intellectual abilities improved grades among lower-achieving pupils. School context helped explain where benefits appeared.

This supports a specific intervention for specific pupils. It does not establish that positive thinking alone reliably improves everyone’s grades.

[A national experiment reveals where a growth mindset improves achievement](https://www.nature.com/articles/s41586-019-1466-y)

## 3. Screen time is a crude explanation of well-being.

Across three datasets with 355,358 adolescents, digital technology use accounted for at most 0.4% of variation in well-being. A statistically detectable association can be small.

This observational result does not establish causation or show that every platform, activity or child faces the same risks. It is a landmark study, not a verdict on today’s apps.

[The association between adolescent well-being and digital technology use](https://www.nature.com/articles/s41562-018-0506-1)

## 4. A useful nudge may stop working when the programme ends.

Researchers compared 54 programmes for 61,293 gym members. During the four-week intervention, 45% of conditions increased visits relative to a placebo control; only 8% did so in the next four weeks.

Measure persistence after support ends. Those percentages count experimental conditions with statistically significant results, not the percentage of people who formed a habit.

[Megastudies improve the impact of applied behavioural science](https://www.nature.com/articles/s41586-021-04128-4)

## 5. Reliable brain–behaviour links can require thousands of people.

A large neuroimaging analysis found that studies linking brain-wide differences to behavioural traits often need samples in the thousands to produce reproducible associations. Small samples can exaggerate the apparent relationship.

Ask how many people were studied and whether the association held in new data. This sample-size lesson concerns brain-wide association studies, not every use of brain imaging.

[Reproducible brain-wide association studies require thousands of individuals](https://www.nature.com/articles/s41586-022-04492-9)

## 6. Changing a belief does not guarantee changing an action.

A climate intervention experiment with 59,440 participants in 63 countries found that different messages worked best for beliefs, policy support and sharing information. None increased the study’s more effortful tree-planting task.

Test the action you care about. A change in a questionnaire answer is a different outcome from sustained behaviour; one experimental task cannot represent all climate action.

[Addressing climate change with behavioral science: A global intervention tournament in 63 countries](https://pmc.ncbi.nlm.nih.gov/articles/PMC10849597/)

An editorial selection of six influential studies published between September 2016 and September 2026, checked on 5 September 2026. “Important” reflects editorial judgment, without an objective ranking or a systematic review of the whole field. We chose findings that change how evidence is interpreted or interventions are designed. Years mark publication, not the origin of each idea. Summaries retain the limits of the cited studies.