Retrieval practice has broad support, but its benefit depends on the alternative, feedback, subject and outcome. Its advantage is clearer against rereading than against other active methods, transfer to new questions is real but conditional, and the mathematics evidence is inconclusive.
Learn / Evidence
Evidence Library
Which behavioral-science findings hold up, and what do they justify doing?
Explore the evidence behind familiar claims about learning, personality, decisions and behavior change. Each evidence review explains the question researchers tested, what later studies changed, and what the findings mean for practical advice. Successful replications sit here beside disputed results and retracted papers.
Choose a reading path
Improve learning
Compare retrieval practice, spaced practice and learning-styles matching: two study methods with broad support and one popular teaching idea with weak support. Together they show what the evidence does, and does not, justify recommending.
Assess an intervention
Start with nudges at scale, then compare if–then plans and growth mindset interventions. In each case, watch three things: the outcome, the setting and the size of the benefit.
Understand a disputed finding
Start with power posing, then read signing first and the 2002 procrastination and deadlines study to follow what later evidence and publication notices changed.
14 evidence reviews
All evidence reviews
The original hormone and risk-taking claims have not held up in replications; a 2026 study, not preregistered, found that posture changed how risk-taking developed over a game, a result that needs independent replication. Posture can shift reported feelings, but studies with a neutral comparison suggest slumping lowers them more than a power pose raises them.
Learning-style matching lacks a strong basis for widespread use. A 2024 meta-analysis found a small benefit, but a 2025 review that pooled four earlier matching meta-analyses put the effect near zero, and a 2026 crossover study with children found no significant benefit.
Waiting has predicted some later outcomes, but the associations depend on the sample and analysis. Two adult follow-ups, at 26 and in mid-life, found little reliable prediction from preschool waiting, and the test cannot identify a proven intervention.
A large project repeated selected tests of links between Big Five personality traits and life outcomes. Most links returned, often with smaller effects; later studies found many trait–outcome links sensitive to control variables, and a study of different traits (agency and communion) in Kenya, Namibia and South Africa found smaller associations. Associations are not causes or reliable forecasts for an individual.
The famous pen task has weak support, but other ways of posing a smile can briefly raise reported happiness. In a 2026 Registered Report, one posed smile lifted happiness in a negative context, but the lift weakened with repetition and all but vanished under threat of a penalty; a brief change in feeling is not lasting wellbeing.
Later tests, including a preregistered direct replication, did not reproduce the sign-first advantage, and PNAS separately retracted the original paper over concerns about its data. Some honesty oaths reduced cheating in a large online tax game, but that is a different comparison.
Psychological Science retracted the paper in September 2026, and a close replication did not reproduce its performance advantage. Neither event settles whether other deadlines help.
Large studies have found both near-zero and positive task-specific effects. Neither result establishes a general store of willpower or makes ordinary fatigue unreal.
Some interventions produce small benefits for defined groups, while strong reviews find little overall benefit. A specific program must justify its time and cost.
Government nudge trials averaged 1.4 percentage points against 8.7 in a published academic sample, and the authors estimate that selective publication, worsened by low statistical power, explains about 70 percent of the gap. A small effect can still be worth its cost when the outcome that matters actually changes.
Numbers people compare before estimating (explicit anchors) shift their estimates reliably in replications. Unrelated background numbers (incidental anchors) did not reliably change consumers' price estimates, and a large anchoring effect on bids shrank sharply when replicated.
If–then plans can help people act, but effects vary and estimates shrink after bias adjustment. Clear cues cannot replace feasible actions and a workable environment.
Spacing has broad support for later retention, including a classroom meta-analysis. The right interval, expanding schedules and branded programs need their own evidence: a large school trial of one spaced-learning package found no exam benefit.
No matching evidence reviews
Try a broader phrase or choose another topic.
How the evidence is assessed
Start with the claim, not its reputation. An experiment about an immediate feeling cannot establish a lasting improvement in performance. A relationship between a personality trait and an outcome cannot, by itself, show how to change that outcome.
Each review compares the original evidence with relevant replications, later studies and research syntheses, and links important findings to their sources. When differences in tasks, samples, outcomes or analysis change the answer, the review says so. These are focused evidence reviews, each answering one question, not exhaustive systematic reviews of every paper on a subject.
There is no universal reliability score here. Asking the right question is more useful, and the right question depends on the claim. For a supported finding, ask how large it is and where it applies. For a disputed finding, ask what was tested differently. For an intervention, ask whether it changes an outcome that matters and whether the likely benefit justifies its cost.
Replication, retraction and misconduct are different
Replication asks whether a finding appears again in new data. A result can come back smaller, appear only under some conditions, or not appear at all. One replication result should not automatically become a verdict on a whole theory. The National Academies' explanation of replication describes why judging consistency depends on the research question and on the uncertainty in the results.
Retraction changes a paper's formal publication status. The reason matters, so read the notice and ask what it changes about that paper's support for a claim. Other, independent evidence may still be relevant. The Committee on Publication Ethics' guidelines cover unreliable findings that arise from honest error as well as from misconduct.
Research misconduct means conduct such as fabrication or falsification. A failed replication alone does not establish misconduct, and a retraction does not automatically identify who was responsible. Any report of a formal misconduct finding must say who made the finding and what it covers.
For definitions, see the glossary. For the wider context of applied behavioral science, continue to Learn.