Feedback improves learning by a medium amount on average, but its effect varies widely with the information it carries, and a substantial share of feedback effects are negative¶

Claim · All claims
Evidence · 2 studies · 2 quant-synthesis r2 · q4 · i2 medium
Currency · newest study 2020 · checked against its sources 2026-10-20 (automated, 2 of 2 entries) · not reviewed by a person

Information provided to learners about their performance or understanding can improve subsequent learning, relative to practice without such information. This page covers the general claim; its strength depends heavily on the form, timing, and content of the feedback and on the learner's stage of expertise.

Subclaims¶

q4 i2 On average, feedback improves student learning by a medium amount (d = 0.48 across 435 studies), but the effects vary widely, and the size depends on how much information the feedback carries: reinforcement or punishment d = 0.24, corrective feedback d = 0.46, high-information feedback d = 0.99. → Wisniewski et al. 2020

q4 i2 Feedback interventions improve performance on average (d = .41), but over 38% of the 607 effect sizes were negative, and the authors' theory, tested with moderator analyses across laboratory and field tasks, holds that feedback works less well the more it turns the recipient's attention away from the task and toward the self. → Kluger & DeNisi 1996

q4 i2 A meta-analysis of 435 studies (k=994 effect sizes, N>61,000) finds a medium overall effect of feedback on student learning (d=0.48), but the pooled effect masks large heterogeneity by feedback type and outcome domain — feedback is not a single consistent treatment. → Wisniewski et al. 2020

q3 i2 Across 607 effect sizes, feedback interventions have a moderate positive average effect on performance (d = .41), yet over 38% of the effects are negative. → Kluger 1996

Evidence¶

Wisniewski et al. 2020¶

Wisniewski, B., Zierer, K., & Hattie, J. (2020). The power of feedback revisited: A meta-analysis of educational feedback research. Frontiers in Psychology, 10, 3087. doi:10.3389/fpsyg.2019.03087

q4 · random-effects meta-analysis · i2 · medium effect, d=0.48 · n=435 studies (k=994 effects, N>61,000) · quant-synthesis · r2

A random-effects meta-analysis of empirical studies of feedback on student learning in education, drawn from the studies behind earlier meta-analyses. The overall effect was medium (d = 0.48), but heterogeneity was large, so the authors argue that feedback should not be treated as a single intervention. The most important moderator was how much information the feedback carried: reinforcement or punishment gave d = 0.24 [0.06–0.43], corrective feedback d = 0.46 [0.39–0.55], and high-information feedback, which adds information about self-regulation, d = 0.99 [0.82–1.15]. Effects were larger on cognitive (d = 0.51) and motor-skill (d = 0.63) outcomes than on motivational ones (d = 0.33). 17% of the effects were negative, fewer than the "about a third" the authors note for Kluger and DeNisi (1996).

Kluger & DeNisi 1996¶

Kluger, A. N., & DeNisi, A. (1996). The effects of feedback interventions on performance: A historical review, a meta-analysis, and a preliminary feedback intervention theory. Psychological Bulletin, 119(2), 254–284. doi:10.1037/0033-2909.119.2.254

q4 · meta-analysis · i2 · medium effect, d=.41 · n=607 effect sizes (23,663 observations) · quant-synthesis · r2

A meta-analysis of feedback interventions on task performance in laboratory and field settings, not only in education. On average, feedback improved performance (d = .41), but over 38% of the 607 effect sizes were negative, and sampling error or whether the feedback was positive or negative did not explain this. The authors' feedback intervention theory, tested with moderator analyses across these task-performance settings, proposes that feedback becomes less effective as it pulls the recipient's attention away from the task and toward the self. This is a strong qualification of the general claim.

Studies bearing on this claim¶

What should the feedback after a check say, and who or what should give it? This table gathers the studies behind this claim and the 30 narrower claims that bear on it: 25 distinct studies (8 causal, 8 review, 2 quant-synthesis, 2 associational, 2 design, 2 qualitative, 1 theoretical), q1–q4, 5 of 25 with an effect size coded. One row per study, codes as each entry records them; no effect is averaged and no verdict is drawn. Built by scripts/synthesize_claim.py from eval/synthesis/; a row that looks wrong is fixed on the claim it comes from.

Discussion¶

Feedback is not uniformly beneficial. The broad claim hides substantial failure modes. Feedback that merely reports right or wrong without explaining why [~M], feedback delivered at a moment when the learner can no longer act on it [-M], and feedback that redirects attention to the self rather than the task [-M] have all been associated with null or negative effects in the wider literature. The claim should therefore be read as conditional: feedback improves learning when it gives the learner usable information they did not already have [+S]. Where feedback functions only as praise or reward, its effects are better explained by behaviorism than by information processing — and such effects rarely transfer to learning outcomes [-M].

What the feedback is about matters. Feedback specifying what was wrong and how to improve tends to support learning more than feedback that only evaluates [+M]. Feedback directed at the self ("good job") is generally the least productive focus [-M], and feedback on self-regulation — prompting learners to monitor and adjust their own strategies — may have the most durable effects [+W], though it is also the hardest to deliver well. This aligns with the broader pattern in which active learning improves exam performance: feedback that prompts the learner to do cognitive work outperforms feedback that does the work for them.

Timing is a moderator, not a simple rule. Immediate feedback helps when the content is not yet retrievable or when errors would otherwise be rehearsed and consolidated [+M]. Delayed feedback can help when it gives learners a chance to attempt retrieval first, or when the intervening activity would otherwise be abandoned [~M]. The question for designers is not "immediate or delayed?" but "will the learner still be able to act on this, and will acting on it involve productive effort?"

Working memory and actionability constrain feedback uptake. Feedback that arrives when working memory is already saturated [-M], or that arrives in a volume the learner cannot act on [-M], is effectively lost. Keeping feedback specific, task-focused, and within what the learner can process in the moment — consistent with cognitive load reduction improves learning — is a precondition for any benefit.

Expertise changes what feedback is for. Novices benefit most from feedback on procedures and concepts [+M]; more advanced learners may find the same feedback redundant or even irritating [-M], an instance of the expertise reversal effect. Feedback design should fade in specificity and shift toward self-evaluation as competence grows.

Relationship to formative assessment. Feedback is the primary mechanism through which formative assessment influences achievement [+S], but the two claims are distinct: assessment for learning improves achievement covers the full cycle of eliciting evidence, interpreting it, and adapting instruction. This page should focus on the feedback episode itself, not the whole cycle.

Design implications. Feedback is most likely to improve learning when it is specific and task-focused [+M], delivered close enough to practice that the learner can still revise [+M], kept within working-memory limits, and paired with an opportunity to act on it — for example through action-oriented feedback that names the next concrete step [+W]. Rubrics and exemplars can make feedback criteria visible in advance, so the feedback episode confirms rather than introduces expectations [+W].

Open questions. The relative contributions of task-level, process-level, and self-regulation feedback across domains remain unsettled, as do the boundary conditions for immediate versus delayed delivery. How automated feedback systems should adapt timing and specificity to learner expertise at scale is an active design problem.

Merged from “Feedback improves learning outcomes” (feedback-improves-learning-outcomes): Actionability. Feedback improves learning when it tells learners something they can act on — what was wrong, why, and what to do next. Feedback that merely evaluates (a grade, a score, "good job") gives the learner no next step. This motivates the design principle of action-oriented feedback, which directs attention to the task and the strategy rather than the person.

Focus of the feedback. Feedback directed at the self ("you're so smart") tends to be least useful and can even backfire by diverting attention from the task to the ego; feedback at the task, process, or self-regulation levels is generally more productive. This is a design choice, not a fixed property of the medium.

Timing and delivery. Immediate feedback often helps for procedural skills and error prevention, while delayed feedback can support retention and transfer; the optimal interval likely depends on the task and the learner's expertise [~M]. Similarly, feedback that reduces cognitive load for a novice can become redundant for an advanced learner — a moderation pattern consistent with the expertise reversal effect [~M].

Relationship to assessment. Feedback is the mechanism through which formative assessment exerts its effect: assessment information only improves achievement when it is translated into feedback learners understand and use — see Assessment for Learning and Assessment for learning improves achievement.

Constraints on effectiveness. The claim fails under identifiable conditions: evaluative-only feedback with no actionable content [-S]; self-level feedback that redirects attention to the ego [-M]; feedback that overloads working memory for novices or becomes redundant for experts [~M]; and feedback the learner cannot understand or act on because it is not aligned with the task or the learner's current understanding [-S].

Open questions. The claim as stated is too broad to be testable as written. Future enrichment should decompose it into subclaims specifying feedback type (verification vs. elaborated), timing, source (teacher, peer, automated), and learner expertise, each with its own evidence base.

Merged from “Feedback improves learning, with a medium average effect that varies widely by feedback type” (feedback-use-improves-learning): Feedback is one of the most consistently reported influences on achievement in the learning sciences, but its effects are famously variable: the same feedback can help, do nothing, or harm depending on its content, timing, and how the learner responds. A useful distinction separates feedback about the task (what was wrong and how to improve), about the process (strategies to try), about self-regulation (how to monitor and adjust one's own work), and about the self (praise or criticism of the person). Task- and process-level feedback are the forms most plausibly tied to improved performance; personal praise carries little instructional information and can even undermine engagement when it frames ability as fixed.

The folded page's claim centred on feedback use, not merely feedback delivery; none of the meta-analyses recorded here measures use (revision, re-attempt) separately from delivery. Feedback plausibly improves learning only when

the learner does something with it — revising a solution, re-attempting a task, or adjusting a strategy. This links the claim to self-regulated learning: learners must notice the gap between current and target performance and act to close it. Designers should therefore build in time and structure for feedback uptake (revision cycles, re-submission, immediate re-practice) rather than treating feedback as a terminal event — see Action-oriented feedback for design patterns that specify what the learner should do next.

Key moderators and boundary conditions (the recorded meta-analyses test information content and outcome type, Wisniewski et al. 2020, and task- versus self-focus, Kluger & DeNisi 1996; the rest below are not tested here):

• Timing. Immediate feedback often helps for procedural skills; delayed feedback can aid retention and transfer by spacing re-engagement. The optimal interval likely depends on task type.
• Goal orientation. Feedback is more effective when learners hold mastery goals than when they hold performance goals, where critical feedback can be experienced as a threat.
• Cognitive load. Feedback on complex tasks must be timed and sized so that reading it does not itself overload working memory — see cognitive overload degrades learning and cognitive load theory. Chunked, prioritized feedback is easier to absorb and act on than exhaustive error lists — see Chunking reduces working memory load.
• Expertise. Novices generally need more directive, task-level feedback; advanced learners may benefit more from prompts to self-evaluate, consistent with the expertise-reversal pattern described in expertise reversal effect.

Open questions: how durable feedback effects are over time, how feedback interacts with grading, and which digital feedback formats best support uptake at scale.

Related Claims¶

• Assessment for learning improves achievement — feedback is the central mechanism of formative assessment; the two claims should not double-count the same evidence
• Cognitive load reduction improves learning — feedback that exceeds working-memory capacity cannot be used, however accurate it is
• Chunking reduces working memory load — structuring feedback into digestible units keeps it actionable
• Clear structure improves learning — well-structured feedback is easier to locate, interpret, and act on
• Self-regulated learning — self-regulation-level feedback aims to make learners their own feedback providers
• Intelligent tutoring systems improve achievement over non-individualised instruction — adaptive systems operationalize feedback by tailoring responses to learner performance
• Conflict-based instruction improves science conceptual learning, though staged contradictions helped only learners who reported being confused, and no study isolates disequilibrium as the mechanism — feedback that surfaces discrepancies between performance and goals can trigger productive disequilibrium
• Feedback Addressing Task Improves Learning — a narrower finding that bears on this claim
• Cognitive overload degrades learning — overloaded learners cannot process or act on feedback.
• Expertise reversal effect — the optimal feedback type shifts from directive to self-evaluative prompts as expertise grows.
• Feedback that answers three questions (Where am I going? How am I doing? Where to next?) improves learning — a narrower finding that bears on this claim
• Feedback Enhances Retrieval Practice — related
• Feedback about self as a person is the least effective feedback level; task, process, and self-regulation feedback support learning and combine for greater impact — related
• Formative assessment practices coherent with OpenSciEd are constrained by grading expectations misaligned with formative goals and by the time cost of evaluating and giving feedback on three-dimensional tasks. — related
• Educators see AI as having potential to increase capacity for high-quality formative assessment, which is an operational challenge at scale — related
• Timely feedback on homework increases learning, and computer platforms providing immediate feedback help teachers adapt — a narrower finding that bears on this claim
• Intelligent tutoring systems can improve learning outcomes, particularly with immediate actionable feedback — related
• Feedback type shifts engagement by question type: educator feedback yields more MCQ submissions while AI feedback trends toward more open-ended revisions, with no overall difference — a narrower finding that bears on this claim
• In the expectations meta-analysis, the amount of feedback agents provided had only a meager relation to performance (r = .07) whereas the climate agents created had a strong one (r = .36) — related
• Feedback is more effective for cognitive and physical outcomes than for motivational and behavioral outcomes — a narrower finding that bears on this claim
• Immediate Elaborated Feedback in a formative assessment game significantly improved students' programming knowledge, interest, and problem-solving versus systems without immediate feedback — a narrower finding that bears on this claim
• Hattie's synthesis identifies lesson study, formative evaluation of teaching, discussion of learning, and feedback as instructional factors with high effect sizes on student learning — related
• Feedback has a medium overall effect on student learning (d = 0.48 after outlier removal) with substantial heterogeneity — possibly the same claim (merge candidate)
• Outcome feedback (mere knowledge of results) impedes learning of complex tasks in the multiple-cue probability learning paradigm — a narrower finding that bears on this claim
• Meta-analysis yields a lower feedback effect (d = 0.48) than the Visible Learning meta-synthesis (d = 0.79), with three synthesized values overestimated — related
• The variability of feedback-intervention effects, including negative effects, exceeds sampling error and is robust to exclusion of a single prolific lab — related
• Teachers provided targeted extra support, immediate feedback, and reinforcers to learners with DD — related