Formative assessment is the systematic use of evidence about students’ current understanding to improve the next decisions made in teaching and learning. Its defining feature is what teachers, students, and peers do with the evidence: identify a specific learning need, adapt instruction or practice, and check whether the response helped. A quiz, question, comment, rubric, classroom discussion, or digital dashboard becomes formative through that responsive use. The process is often called assessment for learning. It occurs before, during, and after individual learning activities, while a meaningful opportunity to improve remains. This functional definition follows the account developed by Black and Wiliam (2009) and the historical analysis of Sadler (1989).
The evidence favors formative assessment as a productive component of instruction, while placing real limits on sweeping promises. A 2024 synthesis of 118 K–12 primary studies reported a mean achievement effect of Hedges’ g = 0.25, and g = 0.22 in its U.S. subset; this is an average across diverse interventions, not a guaranteed gain for every classroom Yao et al. (2024). An umbrella review of 13 meta-analyses found generally positive effects but mostly low or very low certainty in the underlying evidence, making implementation details and research design essential to interpretation Sortwell et al. (2024).
The core scientific question is therefore how information about learning becomes better learning. Answering it requires examining the quality of tasks that reveal student thinking, the interpretation of responses, the instruction that follows, and the student’s opportunity to act. It also requires separating learning from grades, confidence, engagement, and completion of an assigned task. The OECD’s analysis of formative assessment and feedback (2025) places learning goals, diagnostic interpretation, adaptive teaching, and learner agency within a single instructional process.
Definition and Scientific Status
Formative assessment is a purpose and a process rather than a proprietary instrument or a particular frequency of tests. The teacher establishes a learning intention, elicits evidence tied to that intention, interprets what the evidence indicates, and selects an educational response. Learners participate by explaining their thinking, comparing current work with criteria, revising, practicing, or requesting help. An identical multiple-choice question may be used formatively when its response changes instruction and summatively when it only contributes to a final mark. The distinction concerns decisions and consequences, not the physical format of the assessment Black and Wiliam (2009).
Formative assessment has a strong theoretical foundation in educational psychology and a substantial, methodologically varied empirical literature. It combines perspectives from cognitive diagnosis, instruction, motivation, self-regulation, classroom interaction, and educational measurement. Its component practices include questioning, short checks for understanding, feedback, peer dialogue, and student self-assessment. Because these activities differ in structure, cost, and causal mechanism, researchers must specify which component was implemented, how consistently, and against which comparison condition Maskos et al. (2025).
Formative, diagnostic, interim, and summative assessment
A diagnostic assessment usually identifies starting knowledge, misconceptions, or support needs before a particular sequence of instruction. Formative assessment uses evidence to guide next steps throughout that sequence. Interim or benchmark assessments occur periodically and may serve monitoring or accountability purposes. Summative assessment documents what has been achieved at a defined endpoint. These categories overlap in actual classrooms: an end-of-unit test can inform later instruction, while a brief entry question can expose readiness. It is the justified interpretation and use of the information that determine the role, and no label guarantees an instructional benefit Black and Wiliam (1998); OECD (2025).
Student screening, norm-referenced identification, psychological diagnosis, and eligibility determination answer different questions and require their own standards. A low score on a formative classroom task identifies a learning problem under particular conditions; it cannot establish a disorder or explain its cause. Assessment for instructional adjustment belongs to educational practice, whereas standardized measurement validity and disorder-specific evaluations have distinct scientific and professional ownership Harris et al. (2022).
Historical Foundations and Theoretical Models
From feedback research to assessment for learning
Work on feedback, mastery learning, and classroom assessment converged during the twentieth century on the educational significance of information acquired before instruction has finished. Sadler’s influential 1989 model emphasized the learner’s understanding of quality, the ability to compare present work with a standard, and the ability to act on the gap. The landmark review by Black and Wiliam broadened this discussion to classroom systems, feedback, self-assessment, and teacher decision-making, helping establish formative assessment as a major research and policy theme Sadler (1989); Black and Wiliam (1998).
The later theory by Black and Wiliam described formative practice as the elicitation, interpretation, and use of evidence by teachers, students, or peers to improve decisions. Its central contribution is the link between evidence and responsive action: information alone has no instructional effect unless it changes a learner’s opportunities, strategies, or understanding. The model also recognizes that students and classmates can become active agents in the cycle, rather than passive recipients of teacher judgments Black and Wiliam (2009).
Learning goals, evidence, and agency
A widely used theoretical structure organizes formative assessment around three questions: what constitutes the intended learning, what is the learner’s current position, and what step should follow. Within this structure, teachers clarify success criteria, engineer discussions and tasks that reveal understanding, and provide feedback that supports progress. Peers become resources for one another, while learners increasingly take responsibility for monitoring and improving their own work. These five strategies constitute a framework for instructional design rather than a claim that any single classroom routine is universally effective Black and Wiliam (2009); Maskos et al. (2025).
Formative assessment also connects with self-regulated learning. Learners must recognize a target, judge their current work against that target, choose a response, and review its effects. Such monitoring develops gradually and depends on disciplinary knowledge and accessible criteria. A child who can select a strategy after receiving feedback has demonstrated a different educational capability from a child who can only repeat a corrected answer Clark (2012).
The Formative Assessment Cycle
Specify what students should understand
The first task is to define the intended knowledge or performance at a useful level of precision. “Understand fractions” is too broad to interpret a single response, whereas “compare two fractions with different denominators and justify the comparison with a number line” identifies observable evidence of understanding. Success criteria should express the important qualities of the work and allow students to recognize relevant improvements. They should be faithful to the disciplinary goal rather than reducing complex knowledge to easily scored fragments Sadler (1989); OECD (2025).
Learning intentions guide the choice of evidence. For conceptual understanding, the assessment must reveal reasoning or explanation. For procedural fluency, it may require accurate independent performance across representative problems. For transfer, the student should apply the idea in a sufficiently different situation. A task that measures recall alone cannot support a strong inference about flexible use. Alignment between intended learning, assessment design, and response is a methodological as well as pedagogical requirement Harris et al. (2022).
Elicit interpretable evidence
Useful questions make student thinking visible. A diagnostic multiple-choice item may offer distractors corresponding to distinct misconceptions; a short explanation can reveal the rule a learner is using; a worked solution can expose a breakdown in a multistep process. Teachers can collect individual written responses before discussion, preventing a few confident voices from being mistaken for the understanding of an entire class. The evidence should identify actionable differences rather than merely sort students into “right” and “wrong” groups OECD (2025).
The quality of an elicitation task depends on its content, language, timing, accessibility, and susceptibility to guessing. A learner may know the concept but misunderstand the question; may reproduce an algorithm without understanding it; or may be influenced by cues supplied by the teacher. A single response often supports several competing explanations. Sampling more than one task, listening to reasoning, and comparing evidence over time help avoid premature conclusions Harris et al. (2022).
Interpret patterns, then decide
Interpretation asks what the response reveals about the learning goal, how certain that inference is, and which response is likely to help. If most students show the same misconception, the next step may be explicit whole-class clarification. If a smaller group needs additional prerequisites, temporary small-group support may be more appropriate. If students demonstrate secure understanding, the teacher can raise complexity, ask for explanation, or move toward transfer. Decision rules should be pedagogical and revisable; they should not automatically attach an enduring label to a student Black and Wiliam (2009); OECD (2025).
Act and check again
The assessment cycle closes when evidence changes instruction or learning activity and the result is examined. Teachers may reteach a concept with a different representation, model a strategy, assign a focused practice task, arrange peer explanation, or offer an extension. Students need an opportunity to revise, practice independently, and demonstrate the target later. Rechecking reveals whether the chosen response was useful and whether new errors have appeared. This repeated loop explains why formative assessment is best understood as part of teaching, not as a separate testing event Education Endowment Foundation (2021).
Classroom Methods That Can Produce Useful Evidence
Diagnostic questions and all-student responses
Short diagnostic questions can be asked orally, on paper, or through response technology. A strong question has a clear intended inference, plausible alternative answers, and a follow-up plan. Teachers may ask every learner to select an answer and explain a reason before showing aggregated results. The class distribution indicates where to investigate; the explanations reveal whether a popular option reflects a shared misconception, a reading difficulty, or random guessing. A question administered without follow-up may still be a retrieval activity, but its specifically formative value depends on how the evidence shapes subsequent learning Black and Wiliam (2009).
Exit tickets and entrance checks
An exit ticket can ask for one explanation, one worked step, or one unresolved question. Its advantage is speed and proximity to instruction. Its limitation is sparse sampling: a correct answer may reflect a prompt still held in short-term memory, while a poor answer may be produced by haste or fatigue. Exit-ticket patterns should guide a subsequent lesson and, where practical, be checked with delayed or independent performance. Merely collecting large numbers of tickets without reviewing them creates administrative work rather than a reliable learning cycle OECD (2025).
Observation, questioning, and classroom dialogue
Teachers can infer how students reason from the explanations and questions exchanged during discussion, provided attention is paid to whose thinking is represented. Waiting for responses, asking students to elaborate, comparing explanations, and recording recurring misconceptions can generate richer evidence than checking correctness alone. However, discussion is socially selective: fluent, confident, and linguistically advantaged students may dominate. Combining talk with individual written or otherwise accessible responses strengthens inference about the class Harris et al. (2022); Maskos et al. (2025).
Worked examples, drafts, and annotated student work
When the target is a multistep procedure or complex written product, the most informative evidence may be the intermediate work. In mathematics, the teacher can inspect the step where a transformation first becomes invalid. In writing, a draft may reveal whether a claim has evidence and whether the explanation establishes its relevance. In science, a model can reveal the relation a student believes holds between variables. Annotation is useful when it identifies a decision the student can make next; exhaustive correction of every surface feature can obscure the principal learning need Sadler (1989); Education Endowment Foundation (2021).
Rubrics, exemplars, and criteria
A rubric can make important dimensions of quality explicit. Exemplars and comparison of contrasting samples can help students learn what those dimensions mean in practice. The main risk is replacing thoughtful judgment with mechanical box checking, especially for open-ended work. Effective criteria focus attention on the quality of reasoning or performance, are introduced before revision, and support discussion of why one example better meets a learning goal. Students need opportunities to apply the criteria to actual work and receive calibration from more knowledgeable feedback Sadler (1989).
Low-stakes quizzes and cumulative checks
Low-stakes quizzing can supply information to teachers and create opportunities for students to retrieve knowledge. These are overlapping but distinct functions: retrieval may aid memory even if no instructional decision changes, while formative assessment involves interpreting results and acting on them. A systematic review of causal evidence in higher education identified low-stakes quizzing as a particularly promising approach, while also finding that the total trial evidence base for feedback and formative assessment in universities remained limited Morris, Perry, and Wardle (2021).
Feedback as a Component of Formative Assessment
Feedback describes information that can help a learner improve relative to a goal; formative assessment encompasses the larger process of obtaining and using evidence. Strong feedback locates a meaningful feature of the current work, indicates why it matters, and points to a practicable improvement. For instance, “your explanation identifies the correlation but has not ruled out a plausible third variable; compare two alternative causal accounts” supplies actionable information that “good effort” or a numerical score does not. The informational content of feedback is an important moderator of its effectiveness Wisniewski, Zierer, and Hattie (2020).
The evidence is substantial but heterogeneous. A meta-analysis encompassing 435 studies, 994 effect sizes, and more than 61,000 learners reported an overall effect of d = 0.48 for feedback on learning, with substantial variation across studies and greater impacts on cognitive and motor than on motivational and behavioral outcomes. These estimates address feedback interventions broadly and cannot be substituted for the distinct mean effect of formative assessment programs Wisniewski, Zierer, and Hattie (2020).
Feedback can also redirect attention unproductively. A classic synthesis of interventions across performance settings found that some feedback interventions worsened performance, despite a positive average. Its diverse settings should not be treated as a direct estimate for K–12 instruction, yet the finding demonstrates why feedback quality, task focus, timing, and learner response must be examined rather than assumed Kluger and DeNisi (1996).
Timing, medium, and opportunity to respond
Immediate feedback may be useful for preventing a procedural error from being rehearsed, while a delay can sometimes support independent checking and more reflective work. The appropriate timing depends on task difficulty, learner knowledge, and whether the student has a concrete chance to act. Written versus oral, digital versus paper, and teacher versus peer are delivery distinctions that alone do not determine benefit. The Education Endowment Foundation recommends prioritizing the principles of effective feedback, making it usable, and planning for students to respond Education Endowment Foundation (2021).
Peer feedback and peer assessment
Peers can help one another detect errors, compare explanations, and refine judgments. Their comments are especially useful when the learning goal and success criteria are understood, feedback is specific, and the task allows meaningful revision. Peer evaluation can also expose misunderstanding if students merely exchange grades or reinforce one another’s mistakes. The teacher remains responsible for quality assurance, modeling, respectful discussion, and equitable participation. University trial evidence suggests that some peer and tutor feedback practices help, but effects are sensitive to implementation Morris, Perry, and Wardle (2021).
Self-assessment and calibrated judgment
Student self-assessment is formative when it produces a defensible judgment and a change in action. Students can compare work with criteria, identify uncertainty, select a revision, and then evaluate the outcome. Confidence ratings alone are weak measures of knowledge when learners are poorly calibrated; a correct answer accompanied by a sound explanation is stronger evidence. With structured practice and feedback, students can become better at monitoring their learning, which links formative assessment to self-regulation Clark (2012); Sadler (1989).
Why Formative Assessment Can Improve Learning
One causal pathway is better instructional targeting. Teachers cannot reliably address a misconception they have not observed. When evidence identifies a specific obstacle and the response tackles that obstacle, time and practice can be allocated more precisely. A second pathway is feedback that helps learners connect current performance with criteria. A third is active monitoring: learners notice when a strategy fails and choose a more promising one. These are theoretically plausible interacting pathways rather than a single mechanism experimentally isolated by every formative assessment study Black and Wiliam (2009); Clark (2012).
Formative assessment may also influence motivation and classroom relationships. Clear, actionable criteria can make improvement more controllable, while a classroom norm in which errors are treated as evidence for instruction may support participation. These effects depend on how grades, comparison, error exposure, and teacher responses are handled. The cognitive achievement evidence is stronger than claims that formative routines reliably produce large gains in well-being, autonomy, or motivation in all settings Wisniewski, Zierer, and Hattie (2020); Sortwell et al. (2024).
What Experimental and Review Evidence Shows
K–12 pooled effects and practical interpretation
The 2024 meta-analysis by Yao and colleagues synthesized 258 effect sizes from 118 primary studies of K–12 formative assessment and reported an overall effect of g = 0.25, with g = 0.22 for the U.S. subset. The broad definition included different assessment formats and intervention packages. A pooled standardized effect summarizes group differences on the outcomes measured; it does not translate mechanically into additional months of schooling, a fixed test-score increase, or an expected benefit for each child. Effects are conditional on the study sample, outcome measure, comparator, and instruction actually implemented Yao et al. (2024).
An umbrella review published the same year included 13 meta-analyses covering K–12 formative assessment. It found effects ranging from trivial to large, but nine of those syntheses received very low and three low certainty ratings under the review’s grading approach. The meta-analyses often covered partly overlapping intervention types and primary studies, which means adding their results is not equivalent to collecting independent trials. This review supports the general plausibility of benefit while limiting strong claims about an optimal universal strategy or a precise average effect for any particular school Sortwell et al. (2024).
A large school-level randomized evaluation
A pragmatic test of scale comes from the Embedding Formative Assessment program evaluated in 140 English secondary schools and approximately 25,000 pupils. The intervention supported teacher learning communities and peer observation over an extended period. Its randomized effectiveness evaluation found an overall advantage equivalent to roughly two months of additional progress on the Attainment 8 examination measure, with a high security rating for that result. The evaluation did not find clear improvements in English or mathematics GCSE attainment when examined separately Education Endowment Foundation (2018).
This is evidence about a sustained professional development and implementation package, not proof that a particular exit ticket, digital polling platform, or feedback template independently caused the difference. Nor does the conversion to “months” provide a universal psychological unit of learning. The trial is especially informative because it studied ordinary school implementation at substantial scale, while the program’s demands on teacher time and local context remain part of any transfer decision Education Endowment Foundation (2018).
Subject-specific evidence
A 2025 systematic review of formative assessment in mathematics identified 45 studies from 2015–2023 and reported mixed cognitive and noncognitive results. The authors highlighted different definitions, measurement choices, study designs, duration, and implementation components; some computer-based assessments in interactive settings appeared promising. Findings in one subject cannot simply be copied into reading, foreign-language learning, laboratory science, or arts instruction because the evidence needed to detect understanding differs substantially Maskos et al. (2025).
Learning progression-based assessment provides another relevant body of research. A systematic review of 59 studies found evidence that progressions can support teachers’ judgments and instructional decisions, but only six of the 23 studies reporting achievement used experimental designs sufficient for stronger causal inference. This distinction matters: evidence that teachers find a progression useful is different from evidence that the progression raises independent student achievement Harris, Adie, and Wyatt-Smith (2022).
University students and adult learners
Evidence cannot be transferred automatically from children to universities. Morris, Perry, and Wardle reviewed causal studies of formative assessment and feedback in higher education and judged the high-quality evidence base limited. Low-stakes quizzes were promising; peer and instructor feedback appeared beneficial under some conditions; praise, grading, and technology-mediated approaches yielded mixed findings. University courses differ in disciplinary demands, student autonomy, scale, and feedback opportunities, making course-specific design and direct evaluation necessary Morris, Perry, and Wardle (2021).
Reading Educational Evidence Without Inflating Claims
A research report should identify the units assigned to conditions, the comparator, how the practice was implemented, who assessed the outcomes, and when learning was measured. Random assignment of schools or classrooms can reduce some selection bias, but cluster designs need appropriate statistical handling. Quasi-experimental comparisons remain vulnerable to unobserved differences, including teacher preparation and student prior knowledge. Before-and-after improvement without a credible comparison cannot establish that the formative assessment intervention caused it Yao et al. (2024); Sortwell et al. (2024).
Outcomes vary in proximity to teaching. A researcher-made test closely aligned with practiced tasks may show a larger effect than an independent standardized assessment. Immediate posttests estimate short-term attainment, while delayed tests better address retention; unfamiliar tasks assess transfer. Surveys of confidence or satisfaction measure different constructs from academic achievement. Intervention studies should report all meaningful outcomes and relevant adverse effects rather than selecting only the most favorable measure Morris, Perry, and Wardle (2021); Harris et al. (2022).
Effect sizes also require careful interpretation. Hedges’ g and Cohen’s d express differences in standard-deviation units under specific assumptions, not a constant percentage of learning. Heterogeneity may reflect different ages, subjects, feedback modes, duration, baseline attainment, instructional materials, or bias. Small-study effects, publication bias, inconsistent intervention definitions, overlapping meta-analytic inputs, missing implementation data, and weak follow-up reduce confidence in general claims. Evidence of an average benefit can coexist with a considerable probability of smaller, absent, or unfavorable effects in a particular application Sortwell et al. (2024).
Learning Progressions and Diagnosing Misconceptions
A learning progression describes a hypothesized sequence of increasingly sophisticated understanding within a domain. It can help teachers choose assessments and interpret a response as a step along a conceptual path, rather than a simple pass or fail. However, progressions are models that require validation: actual learners may move through different routes, a response may fit more than one level, and the same task may function differently across populations. Their usefulness in formative practice is better supported than strong claims about raising standardized achievement in every context Harris et al. (2022).
Misconception diagnosis is similarly provisional. If a student selects a distractor that was designed to represent a particular mistaken rule, the response makes that explanation more plausible, not certain. An interview, worked explanation, or second item can discriminate between misconceptions and language or attention errors. Strong formative interpretation keeps alternative explanations in view and responds with instruction that can reveal whether the initial diagnosis was correct OECD (2025).
Equity, Inclusion, and Accessible Assessment
Formative assessment can reveal who is being left behind by an instructional sequence, yet unequal opportunities to show knowledge can distort the evidence. Oral-only participation may disadvantage some multilingual students, students using augmentative communication, or students uncomfortable speaking in large groups. Fast timed responses can overemphasize processing speed. A culturally unfamiliar prompt can assess background knowledge that the task did not intend to measure. Teachers can offer multiple appropriate response modes, review whether wording is comprehensible, and distinguish access barriers from evidence of the target knowledge OECD (2025).
Flexibility should preserve the construct being assessed. If the goal is mathematical proportional reasoning, an accessible response format can reveal that reasoning without unnecessary written-language demand. If the goal is formal writing, written expression is part of the construct and calls for different adjustments. Universal instructional accessibility and disability-related individualized accommodations may intersect, but they operate under distinct pedagogical and professional requirements. A teacher should avoid turning formative error patterns into diagnostic conclusions Harris et al. (2022).
Evidence about distributional effects is promising but uneven. The Embedding Formative Assessment trial reported indications of larger benefits among lower-attaining pupils, yet the subgroup analyses had lower certainty than the overall estimate. The 2025 mathematics review also identified possible advantages for lower- and medium-performing students and learners from lower socioeconomic backgrounds, while emphasizing the heterogeneity and incompleteness of that evidence. Equity claims therefore require pre-specified subgroup analysis, sufficiently large samples, and transparent reporting rather than assuming that every assessment routine closes achievement gaps Education Endowment Foundation (2018); Maskos et al. (2025).
A respectful error culture matters because formative evidence is often public or personally revealing. Anonymized aggregate responses can allow a teacher to discuss misconceptions without exposing individuals. Private feedback and opportunities to revise can support constructive risk-taking. Teachers should distinguish evaluating a piece of work from evaluating the person who produced it, and avoid permanent ability labels derived from momentary performance. These recommendations follow the function of responsive assessment and broader feedback evidence, while their specific social effects require context-sensitive evaluation Education Endowment Foundation (2021); Clark (2012).
Digital Formative Assessment and Artificial Intelligence
Learning platforms can collect responses rapidly, display misconception patterns, and supply feedback at scale. Their educational value depends on item validity, useful interpretation, appropriate instruction, and evidence of learning outcomes. A dashboard that highlights the most frequently missed question is a representation of performance, not an explanation of the underlying misconception. Adaptive practice may be useful when it offers genuine opportunities to respond, checks independently produced answers, and records progress toward a clear goal Maskos et al. (2025).
The research on general digitally delivered feedback should also be interpreted at the level of its own intervention class. Digital media can improve feedback speed and accessibility, but delivery technology does not fix poor task selection or inaccurate comments. Reviews in higher education report mixed results for technology-based feedback, and broader digital-prompt synthesis shows that the content and timing of prompts, learner group, setting, and publication-bias adjustment can change apparent effects Morris, Perry, and Wardle (2021); Thomann and Deutscher (2025).
Generative AI may draft quiz items, synthesize free-text responses, propose explanations, or provide conversational hints. Such uses require rigorous human review: generated questions can contain incorrect keys or ambiguous distractors, and a plausible explanation can conceal an error. Answers completed by an AI system do not by themselves show what the student can do independently. An educational evaluation should distinguish student-produced reasoning, immediate task performance with support, later unaided retention, transfer, teacher workload, and error rates. Evidence about general AI achievement or chatbot interaction should not be imported as proof of formative assessment efficacy OECD (2025); Morris, Perry, and Wardle (2021).
Digital formative assessment also raises questions of privacy and fairness. Learner responses can reveal disability-related difficulties, language proficiency, academic insecurity, and sensitive personal circumstances. Systems should minimize identifiable data, clarify who can access records, provide defensible retention and deletion rules, and avoid using speculative automated labels as fixed judgments about potential. Algorithmic group comparisons demand scrutiny for differential measurement quality. Technology procurement is therefore a decision about pedagogy, measurement, governance, and learner rights together, not a substitute for teacher judgment OECD (2025).
Teacher Expertise and Implementation Quality
Formative assessment asks teachers to identify important concepts, anticipate plausible misconceptions, notice relevant responses, and select a fitting instructional action in real time. These abilities require subject-matter knowledge, task-design skill, and opportunities for reflective practice. The same visible routine can be superficial or intellectually demanding depending on the question and how the response is used. Classroom implementation is therefore a substantive treatment component that should be documented in intervention research Black and Wiliam (2009); OECD (2025).
Professional development should include practice designing tasks, studying student work, rehearsing responsive teaching, and receiving useful feedback on implementation. A 2025 systematic review of 23 studies of professional development in formative assessment found heterogeneous programs, measures, and limited use of intervention-control comparisons. Such evidence constrains claims about one universally best training format. The large Embedding Formative Assessment evaluation shows that a more sustained school-wide program can achieve measurable gains, but it does not settle which component is indispensable von Hagen et al. (2025); Education Endowment Foundation (2018).
Schools should also consider opportunity costs. Every added quiz, marking cycle, feedback platform, or reporting demand consumes instructional and planning time. A narrow set of high-information questions followed by meaningful action may be more valuable than frequent low-information data collection. Sustainability involves protecting time to interpret evidence and adapt lessons, rather than rewarding the number of assessments recorded. The emphasis on responsive practice over paperwork is consistent with professional guidance and the empirical attention to implementation Education Endowment Foundation (2021).
A Worked Classroom Example: Comparing Fractions
A middle-school class is learning to compare fractions by reasoning about magnitude. The teacher sets an explicit goal: students should be able to justify which of two fractions is larger when numerators and denominators differ, using a number line or an equivalent representation. Before new instruction, every student answers a question comparing 3/5 and 4/7 and records a brief explanation. Several choose 4/7 because both the numerator and denominator are larger. The responses reveal a plausible misconception but do not yet show whether students applied a faulty rule or simply guessed.
The teacher then asks a second, independent question comparing 2/3 and 3/5, alongside a number-line illustration. Students explain how they decide whether each value is above or below one-half. The teacher notices two groups: students who compare whole-number parts independently and students who can reason correctly with benchmarks but make procedural calculation mistakes. Rather than assigning everyone the same practice, the teacher models fraction magnitude and benchmarks for the first group and provides calculation checks for the second. Students then complete a new comparison without a model.
Two days later, the class revisits the same conceptual target using unfamiliar pairs of fractions and requires a written justification. A further question asks students to explain why comparing numerators alone fails in some cases. The delayed work helps distinguish temporary performance after explanation from more durable understanding. The teacher uses the new responses to decide whether another representation or additional practice is needed. Each stage has an identifiable learning goal, evidence, interpretation, action, and subsequent check; that is what makes this sequence formative Black and Wiliam (2009); Sadler (1989).
The example is an instructional illustration, not an experimentally validated lesson package. Its value is to make the decision process visible. To test its effectiveness, researchers would need a credible comparator, sufficient classrooms, measures of independent and delayed learning, and fidelity information. A teacher’s impression that students understood the lesson is useful feedback about practice but cannot establish a causal effect Yao et al. (2024).
Common Failure Modes and How to Improve Them
The most common design failure is collecting evidence with no corresponding instructional decision. A second is using a question too vague or poorly aligned to identify what needs attention. A third is substituting scores, generic praise, or extensive marking for information a learner can act upon. A fourth is concluding that the class understands a topic because a vocal subset answers correctly. Teachers can respond by preparing a small set of diagnostic tasks, sampling every student, and planning alternative next steps before collecting the responses Black and Wiliam (2009); Education Endowment Foundation (2021).
Other failures arise when assessments become high stakes or feel punitive. Students may conceal uncertainty, compete for public approval, or focus on point accumulation. When every low-stakes check is converted into a grade, teachers can lose access to candid evidence of misconceptions. A design can instead use selected ungraded opportunities, explicit learning goals, private reflection, structured revision, and occasional independent checks for durable learning. These moves follow formative principles; their success still depends on subject, age, institutional constraints, and implementation Clark (2012); OECD (2025).
Implications for Educational Practice and Policy
For teachers, the operational priority is to connect each important learning goal to a form of evidence that can alter instruction. Start with a misconception that matters, design a task that can reveal it, anticipate the possible response patterns, and allocate time for a specific follow-up. Student work should be revisited later to assess whether the response improved independent performance. This method makes formative assessment manageable and intellectually precise OECD (2025).
For schools, the priority is to develop professional judgment and learning-focused routines rather than to mandate a fixed volume of data collection. Teacher collaboration can improve diagnostic questions and interpretation across classrooms. Any school-wide program should be evaluated through implementation records, opportunity costs, learner experience, and independent attainment measures. The Embedding Formative Assessment trial provides one credible example of a sustained model at scale, while current reviews make clear that its exact impact cannot be assigned automatically to every adoption Education Endowment Foundation (2018); von Hagen et al. (2025).
For researchers and policymakers, the priority is stronger causal identification and more informative reporting. Trials should document what teachers actually did with evidence, pre-register important outcomes, include meaningful comparison practices, consider clusters and missing data, and measure delayed learning as well as immediate scores. Replication in different ages, subjects, languages, and institutional settings is needed. The scientific conclusion is that formative assessment offers useful ways to improve educational decisions, with the size and distribution of learning gains determined by the assessment design, the response it triggers, and the conditions in which it is implemented Sortwell et al. (2024); Maskos et al. (2025).
Frequently Asked Questions
Is formative assessment the same as a quiz?
A quiz can be part of formative assessment, but its status depends on use. When quiz answers help a teacher identify misconceptions, change instruction, and check progress, the quiz serves a formative purpose. When it only records a grade, it is an assessment of attained performance. Low-stakes quizzes can also have separate benefits as retrieval practice, even without changes to teaching Morris, Perry, and Wardle (2021).
Does formative assessment always improve test scores?
No single routine has a guaranteed result. Across diverse K–12 studies, the average effect is positive and typically modest, but intervention effects vary. The quality of the evidence and the outcome measure matter; program effects should be interpreted in relation to their comparator and implementation Yao et al. (2024); Sortwell et al. (2024).
How often should teachers assess formatively?
Frequency should follow the demands of the learning goal and the usefulness of the resulting information. Several brief checks can be appropriate during a challenging sequence, whereas a more complex task may require time for students to produce interpretable work. The critical design decision is whether teachers and learners have time and means to respond to the evidence OECD (2025).
Can formative assessment replace summative assessment?
They serve complementary purposes. Formative assessment informs improvement while learning is underway. Summative assessment provides a defensible record of achievement at a defined point, often for reporting or certification. Evidence collected for one purpose must be evaluated before being used for another, especially when the stakes or interpretation change Black and Wiliam (2009).
Is AI-generated feedback reliable enough for formative assessment?
Reliability depends on content accuracy, the type of task, error detection, and teacher oversight. Generated comments should be checked for correctness and instructional relevance, and students’ later independent learning should be measured. Automatically completing a task or supplying an explanation is not equivalent to demonstrating student understanding. Privacy and access require explicit safeguards OECD (2025).
Related Articles
The published research chapter Cluster Learning and Educational Psychology situates formative assessment among learning theory, metacognition, instructional design, and educational evidence. Retrieval Practice: Why Testing Yourself Improves Memory explains memory effects of testing, a mechanism conceptually related to but separate from evidence-responsive teaching. Spaced Repetition: How the Spacing Effect Supports Long-Term Memory addresses retention over repeated learning opportunities.
Active Learning: What Research Shows About Student Achievement — Evaluate active learning, participation methods, experimental comparisons, and subject-specific evidence.
Effective Feedback for Learning: Timing, Content, and Outcomes — Compare task and process feedback, timing, self-regulation, and circumstances where feedback backfires.
Social-Emotional Learning: Programs, Skills, and Academic Outcomes — Evaluate social-emotional learning programs, taught skills, classroom outcomes, academic performance, and the quality of school-based evidence.
Cooperative Learning and Peer Instruction: Evidence and Conditions — Assess structured collaboration, peer explanation, accountability, and effects on achievement and engagement.
References
Black, P., and Wiliam, D. (1998). Assessment and classroom learning. Assessment in Education: Principles, Policy and Practice, 5(1), 7–74. https://doi.org/10.1080/0969595980050102
Black, P., and Wiliam, D. (2009). Developing the theory of formative assessment. Educational Assessment, Evaluation and Accountability, 21, 5–31. https://doi.org/10.1007/s11092-008-9068-5
Clark, I. (2012). Formative assessment: Assessment is for self-regulated learning. Educational Psychology Review, 24, 205–249. https://doi.org/10.1007/s10648-011-9191-6
Education Endowment Foundation. (2018). Embedding Formative Assessment: Evaluation and project findings. https://educationendowmentfoundation.org.uk/projects-and-evaluation/projects/embedding-formative-assessment
Education Endowment Foundation. (2021). Teacher Feedback to Improve Pupil Learning: Guidance report. https://educationendowmentfoundation.org.uk/education-evidence/guidance-reports/feedback
Harris, L. R., Adie, L., and Wyatt-Smith, C. (2022). Learning progression–based assessments: A systematic review of student and teacher uses. Review of Educational Research, 92(6), 996–1040. https://doi.org/10.3102/00346543221081552
Kluger, A. N., and 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. https://doi.org/10.1037/0033-2909.119.2.254
Maskos, K., Schulz, A., Oeksuez, S. S., and Rakoczy, K. (2025). Formative assessment in mathematics education: A systematic review. ZDM – Mathematics Education, 57, 679–693. https://doi.org/10.1007/s11858-025-01696-x
Morris, R., Perry, T., and Wardle, L. (2021). Formative assessment and feedback for learning in higher education: A systematic review. Review of Education, 9(3), e3292. https://doi.org/10.1002/rev3.3292
OECD. (2025). Using formative assessment and feedback. In Unlocking High-Quality Teaching. https://www.oecd.org/en/publications/2025/04/unlocking-high-quality-teaching_808bd876/full-report/using-formative-assessment-and-feedback_189fb6dc.html
Sadler, D. R. (1989). Formative assessment and the design of instructional systems. Instructional Science, 18, 119–144. https://doi.org/10.1007/BF00117714
Sortwell, A., Trimble, K., Ferraz, R., Geelan, D. R., Hine, G., Ramirez-Campillo, R., Carter-Thuiller, B., Gkintoni, E., and Xuan, Q. (2024). A systematic review of meta-analyses on the impact of formative assessment on K–12 students’ learning: Toward sustainable quality education. Sustainability, 16(17), 7826. https://doi.org/10.3390/su16177826
Thomann, H., and Deutscher, V. (2025). Scaffolding through prompts in digital learning: A systematic review and meta-analysis of effectiveness on learning achievement. Educational Research Review, 47, 100686. https://doi.org/10.1016/j.edurev.2025.100686
von Hagen, A., Balbi, A., Bonilla, M., and Arbildi, C. (2025). What do we mean when we talk about teacher professional development in formative assessment? A systematic review. International Journal of Educational Research. https://doi.org/10.1016/j.ijer.2025.102586
Wisniewski, B., Zierer, K., and Hattie, J. (2020). The power of feedback revisited: A meta-analysis of educational feedback research. Frontiers in Psychology, 10, 3087. https://doi.org/10.3389/fpsyg.2019.03087
Yao, Y., Amos, M., Snider, K., and Brown, T. (2024). The impact of formative assessment on K–12 learning: A meta-analysis. Educational Research and Evaluation, 29(7–8), 452–475. https://doi.org/10.1080/13803611.2024.2363831