Formative Assessment in the Classroom: What Actually Works

Most classrooms run formative assessment every day and still cannot tell which students actually understand the material. A poll, a thumbs up, an exit ticket: these generate data, but the data answers a narrower question than instructors think it does. It tells you whether a student can recognize a correct answer in the moment, not whether they hold a working model of the concept.

Paul Black and Dylan Wiliam's foundational review of classroom assessment research found effect sizes between 0.4 and 0.7 for formative assessment, among the largest of any documented educational intervention, spanning students from kindergarten through university undergraduates. That is a strong case for the practice. It is not, however, a case for any particular implementation of it. The size of the effect depends entirely on what the assessment requires the student to do.

What the Research Actually Supports

Black and Wiliam's review found that the quality of formative feedback mattered more than its frequency. Feedback tied only to a grade or a score did little to change learning; feedback that explained the specific gap and what to do about it changed outcomes. A later meta-analysis on writing instruction found that teacher-delivered feedback produced a larger effect than peer or self-assessment, with computer-generated feedback trailing furthest behind, a gap that traces directly to specificity: a graded rubric response from a real reader catches a misunderstanding a generic prompt does not.

This is the part most classroom tools skip. A multiple-choice poll can tell an instructor that 40 percent of the class picked the wrong answer. It cannot tell them what those students actually believe, which means the instructor is left guessing at the remediation.

Exit Tickets and the Limits of the Quick Check

Exit tickets are a reasonable starting point for assessment in the classroom, and they earned their place in the formative assessment toolkit for a real reason: they are fast, low stakes, and give instructors a same-day read on whether a lesson landed. Their limitation is structural, not a matter of execution. A one or two question exit ticket samples a narrow slice of a topic, and a correct answer on that slice does not rule out a misconception sitting just outside it.

Why exit tickets fall short of measuring understanding

The same is true of cold calling, hand signals, and most digital polling. Each gives a snapshot of recall under low pressure. None of them require the student to construct an explanation, which is the step where gaps in understanding tend to surface.

Why Explanation Outperforms Recognition

The research consensus on this point is consistent: assessment that requires a student to generate an explanation, rather than recognize one, produces a more accurate read on understanding. This is not a new idea in learning science. It is the same principle behind the Feynman Technique, which holds that a person who can teach a concept clearly, in their own words, to someone who does not already know it, has demonstrated something multiple choice cannot capture.

the Feynman Technique as an assessment instrument

A Frontiers in Education review of feedback processing found that students benefit most when feedback forces them to actively reconcile what they believe with what is correct, rather than passively receive a correction. Explanation-based assessment does this naturally: the student has to surface their reasoning before anyone can correct it.

What This Means for Higher Education Assessment

University instructors face a version of this problem at scale. A lecture hall of 200 students cannot be individually interviewed every week, so most assessments in education at the higher education level default to scalable formats: quizzes, polls, problem sets. These are reasonable compromises given the constraints, but they inherit the same limitation as the exit ticket. Scale and depth have historically traded off against each other.

This is the gap Axiom Flow is built to close. Unlike a standard formative assessment platform that monitors recall through quizzes and polls, Axiom Flow measures whether a student can explain and defend what they understand, and does it at a scale a single instructor could not manage through individual interviews. The platform's AI student, Sam, starts with a configurable set of misconceptions generated by Atlas, Axiom Flow's assessment designer. The student's task is to teach Sam until those misconceptions are corrected. Sam then takes a constrained exam using only what it was taught, and Atlas scores the result.

The teaching step is not preparation for the assessment. It is the assessment. This is what assessment for learning looks like when implemented directly: the act of explaining the concept is the same act that produces the measurement, rather than a study step that precedes a separate test.

what assessment for learning means in practice

Building Formative Practice That Holds Up

Instructors do not need to abandon exit tickets, polling, or quick checks. These tools serve a purpose: they are fast, they are low stakes, and they keep a lesson responsive in real time. The mistake is treating their output as evidence of mastery rather than evidence of recognition.

the difference between formative and summative assessment

A more complete formative practice pairs fast recognition checks with at least one explanation-based assessment per unit, the kind that asks a student to teach, defend, or reconstruct a concept rather than select it from a list. That single addition closes most of the gap between what classroom formative assessment currently measures and what the underlying research says it should be measuring.