How Dynamic Assessment Works | Measure What Changes When the Learner Receives Structured Help

eduKateSG Learning Node Series · 0234

A learner gets the question wrong. Then you give one carefully chosen hint, and the learner solves three new questions correctly.

What did the first wrong answer actually tell you?

A conventional assessment often tries to hold help constant or remove it entirely so performance can be compared under standard conditions. That is useful when the job is to estimate what the learner can currently do without extra assistance. It is less complete when the decision is about how the learner responds to instruction.

Dynamic assessment works by deliberately introducing structured help—prompts, feedback, mediation or short teaching episodes—inside the assessment process, then studying how much and what kind of support changes performance. The learner is not simply measured at one unsupported point. The assessment asks how performance moves when the conditions change in a controlled way.

That makes dynamic assessment powerful and easy to overclaim. Responsiveness to help can reveal information that a static score hides. It does not automatically reveal a permanent “learning potential,” prove the cause of the original difficulty, or guarantee that the improved performance will survive once the help disappears.

The 50-second read

• Static assessment asks what the learner can do under specified conditions now.
• Dynamic assessment asks what changes when carefully structured assistance is introduced.
• A common design is test → teach or mediate → retest.
• Another design uses graduated prompts, from the least informative hint to more explicit assistance.
• The amount and type of help required can become part of the evidence.
• Improvement during mediation is not automatically durable learning.
• A fair dynamic assessment needs a defined prompt sequence, stopping rules and comparable tasks.
• If different assessors improvise different help, the result can become a measure of the assessor as much as the learner.
• Dynamic evidence is especially useful when an unsupported score cannot distinguish lack of opportunity, unfamiliarity, weak strategy selection and deeper knowledge gaps.
• Delayed and changed-condition checks are essential if the interpretation is about learning rather than temporary supported performance.
• Research suggests dynamic approaches can add useful information in some domains and populations, but effects and validity evidence vary by age, task, language background and design.
• The safest conclusion is not “this learner has high or low potential.” It is “under these conditions, this learner responded to this form of help in this way.”

Canonical owner boundary

This node owns assessment in which assistance is deliberately varied so the assessor can study responsiveness to mediation, prompts or brief teaching. How Formative Assessment Works owns the broader use of evidence to adjust ongoing learning. How Scaffolding Works owns temporary instructional support. How the Assistance Dilemma Works owns how much help to provide during learning. This article asks a narrower assessment question: what can we learn by making the help itself part of the measurement design?

1. A wrong answer contains more than one possible story

Suppose a student cannot solve 3(x + 4) = 21.

The failure could reflect several different problems: the distributive structure is unfamiliar; inverse operations are insecure; the learner does not know where to begin; the notation is visually confusing; the learner knows the method but fails to retrieve it; or the learner has never seen this form despite understanding equivalent mathematics in another representation.

A static score records the failure. It does not identify which explanation generated it.

A dynamic sequence can change one condition at a time. “What operation is happening to the bracket?” is different help from “expand the bracket first,” which is different again from showing 3x + 12 = 21. The point is not simply to rescue the answer. It is to see which intervention changes the learner’s route.

2. Test–teach–retest asks whether brief instruction changes later performance

One classic dynamic design begins with an initial assessment, introduces a defined teaching or mediation phase, then uses a post-assessment with new but related tasks.

The pretest establishes current performance. The mediation gives the learner an opportunity to acquire or reorganise a strategy. The posttest asks whether the new performance appears after that intervention.

The design becomes more informative when the retest is not an exact repeat. If the learner merely remembers the worked answer, improvement can be mistaken for learning. Parallel tasks, altered numbers, changed surface features or a transfer item help test whether something more general changed.

3. Graduated prompting measures the path through help

A second approach embeds a prompt ladder inside each task. The assessor begins with minimal assistance and increases support only when needed.

1. Orientation: “Read the question again. What is it asking you to find?”
2. Attention cue: “Which part of the equation is acting on the unknown first?”
3. Strategic hint: “Try undoing the multiplication around the bracket.”
4. Partial step: “Divide both sides by 3 first.”
5. Model: show the first transformation, then ask the learner to continue.

The precise ladder depends on the domain. What matters is that the levels are defined in advance strongly enough that “needed three prompts” has interpretable meaning. A casual hint from one teacher and a near-solution from another cannot be treated as the same amount of help merely because both were counted as one prompt.

4. Help amount and help type are different variables

Two learners may each require two prompts, yet the prompts do different jobs.

One learner may need help orienting attention but then execute independently. Another may understand the goal yet need the missing concept supplied directly. Counting prompts alone compresses this distinction.

A stronger record notes the function of the help: attention, representation, strategy selection, factual prerequisite, procedural step, error detection or self-monitoring. These categories are themselves hypotheses and need consistent definitions.

5. Learning-potential scores are constructed, not discovered

Dynamic assessment research has used several quantities: posttest gain, number of prompts required, maximum prompt level, efficiency of improvement, or combinations of initial performance and responsiveness.

None is automatically “the” learning potential. Each operationalisation privileges a different aspect of modifiability. A learner who begins high has less room to improve. A learner who begins low can show a large gain partly because the scale offers more headroom. Prompt counts depend on item difficulty and prompt design.

A useful report names the observed quantity instead of giving it a grander psychological label than the design can support.

6. A worked example: same pretest score, different responsiveness

Consider two hypothetical students on four novel ratio problems. Both solve one of four independently.

Student A fails the second problem. After the cue “find the value for one unit first,” she solves it, then solves two new unit-rate problems without further help.

Student B receives the same cue and remains stuck. A more explicit worked step is provided. He completes the item but fails the two new examples when the prompt is removed.

The static score was identical: 1/4. The dynamic evidence is different. Student A’s difficulty may have involved strategy access or task framing; Student B may need deeper instruction, more practice or a different representation. These remain hypotheses until later evidence supports them.

7. The posttest should separate supported performance from learning

Improvement while the assessor is still prompting is expected. The harder question is whether performance remains improved when the prompt is absent.

Use at least three distances from the intervention when the decision matters:

• Immediate independence: a new item immediately after help.
• Changed condition: a problem with altered surface features, representation or context.
• Delayed return: another opportunity after enough time has passed for simple short-term carryover to weaken.

The stronger the claim about learning, the more important these distances become.

8. Dynamic assessment is not simply formative assessment with more hints

Formative assessment is defined by the use of evidence to improve ongoing teaching and learning. Dynamic assessment is distinguished by making mediation or assistance part of the assessment procedure itself.

A teacher can use a formative quiz without providing help during the quiz. A dynamic assessment can be used diagnostically even when it is not embedded in a full formative assessment cycle. In practice the two can interact, but their defining jobs are different.

9. Dynamic assessment is not ordinary tutoring either

In tutoring, the goal is usually to help the learner improve. The tutor can change explanations freely, choose examples opportunistically and spend more time where needed.

In dynamic assessment, the support must also generate interpretable evidence. If the assessor improvises without recording what changed, the learner may still benefit, but the procedure becomes difficult to compare across learners or occasions.

Teaching flexibility and measurement comparability pull in different directions. Good dynamic designs make that trade-off explicit.

10. Standardisation does not require robotic mediation

Dynamic assessment is sometimes described as either standardised or interactive, as though one must choose between a script and meaningful human mediation.

A better distinction is what must remain stable for the interpretation. The prompt sequence, decision rules, target skill and scoring can be specified while allowing ordinary conversational delivery. More clinical or research-focused approaches may deliberately permit greater assessor responsiveness, but then the result should be interpreted accordingly.

The more consequential the comparison between learners, the stronger the need to document what assistance each actually received.

11. The assessor can become part of the measured system

If one assessor waits longer, paraphrases more skilfully or gives warmer encouragement, learners may respond differently. This is not a reason to remove human interaction. It is a reason to recognise assessor variation as part of the evidence system.

Training, examples of each prompt level, moderation and records of deviations can reduce uncontrolled variation. In research, assessor effects may need explicit modelling.

A score about responsiveness should not silently become a score about which assessor happened to conduct the session.

12. Responsiveness can depend on prior familiarity with help

A learner accustomed to verbal hints may use them efficiently. Another may have been trained to wait for full demonstrations. A bilingual learner may understand the concept but need the prompt expressed differently. A learner unfamiliar with think-aloud interaction may appear less responsive because the assessment format itself is new.

Dynamic assessment can reduce some disadvantages of purely static testing by observing change under support, but it does not automatically remove cultural, linguistic or experiential differences.

Fairness still requires task review, language analysis, comparable opportunity and cautious interpretation.

13. What current research can and cannot say

Dynamic assessment has a substantial research tradition. Swanson and Lussier’s 2001 meta-analysis examined dynamic assessment across multiple studies and contributed to the evidence base on modifiability measures and prediction.

A more recent example is Wood, Biggs and Molnar’s 2024 meta-analysis of dynamic assessments of word reading skills in diverse school-age children. Across 35 studies and 192 effect sizes, the authors found evidence supporting dynamic word-reading assessment, with especially encouraging findings in younger children. They also emphasised limits in the evidence base, including relatively few bilingual samples for some comparisons.

These results do not justify a universal claim that dynamic assessment is always superior to static assessment. They support a narrower conclusion: in some domains and populations, responsiveness to structured support can contribute useful validity evidence beyond an unsupported performance score.

14. Prediction is different from explanation

Suppose responsiveness during a dynamic assessment predicts later achievement better than a pretest alone. That is useful. It does not prove why.

The dynamic measure may capture strategy flexibility, prior instructional opportunity, language comprehension, attention to feedback, motivation, familiarity with mediation or some combination. Predictive usefulness and causal explanation are different achievements.

A responsible report can say what the measure predicts without pretending it has isolated the psychological mechanism.

15. The zone-of-proximal-development idea is influential but not a scoring shortcut

Dynamic assessment is often connected to Vygotskian ideas about the difference between independent performance and performance possible with assistance.

The connection is conceptually useful: current unsupported performance may underestimate what becomes possible through interaction. But a specific prompt score should not be treated as a direct numerical reading of a theoretical “zone.”

Theory motivates what to observe. The assessment design still has to establish what its actual measurements mean.

16. Group dynamic assessment changes the social unit

Some research extends dynamic assessment to groups, where prompts or mediation respond to collective learner needs. A 2023 EFL study, for example, explored group dynamic assessment in a language-learning setting.

Group designs can be efficient and instructionally realistic. They create additional interpretive problems: not every learner receives the same effective exposure, some learners benefit from peers’ responses, and a public prompt may teach students who did not initially need it.

The unit of mediation must match the unit of the conclusion.

17. A tutor can use dynamic logic without pretending to run a formal dynamic assessment

Small-group tutoring can borrow the logic safely:

1. Give one unfamiliar task without help.
2. Record the first valid step and the point of failure.
3. Offer the smallest pre-planned hint likely to discriminate between competing explanations.
4. If needed, increase help one level at a time.
5. After the learner succeeds, remove the help.
6. Use a fresh problem with changed surface features.
7. Return later to see whether the route survives.
8. Record what kind of help changed performance rather than merely writing “understood after explanation.”

This is an instructional diagnostic routine, not a validated clinical or psychometric instrument. Its value is in improving the next teaching decision.

18. Cross-domain comparison: an engineering load test

An engineer may learn more about a structure by applying controlled loads than by looking at it at rest. The response under changing load reveals properties that a single static observation cannot.

Dynamic assessment has a similar logic: change the support condition deliberately and observe the response.

The analogy stops there. A learner changes during the assessment. The “test” can become instruction. That reactivity is not noise to eliminate; it is part of what dynamic assessment is designed to observe.

19. Cross-domain comparison: debugging by controlled intervention

When software fails, a good debugger does not merely rerun the same broken state. It changes one condition: inspect an input, bypass a component, provide a known value, then observe what changes.

A graduated prompt performs an analogous discrimination job. If a representation cue immediately restores performance, that is different evidence from a case in which the complete procedure must be modelled.

Again, people are not programs. The comparison is useful only for the logic of controlled intervention.

20. Failure mode: improvement equals mastery

The learner succeeds with support, so the assessor records the skill as mastered.

Repair: separate supported success, immediate unsupported success, transfer and delayed retention.

21. Failure mode: every assessor invents a different hint ladder

Two learners both receive “two prompts,” but one receives vague encouragement while another receives half the solution.

Repair: define prompt levels by information content and function, train assessors, and document deviations.

22. Failure mode: the posttest is the taught item with different numbers

The learner remembers the procedure exactly and appears to have generalised.

Repair: include tasks that preserve the target relation while changing surface features, representation or context.

23. Failure mode: low responsiveness becomes a learner identity

A learner needs substantial help in one session and is labelled “low learning potential.”

Repair: report the observed interaction: task, assistance, amount, response and date. Reassess with alternative mediation and later evidence before making broad claims.

24. A practical dynamic-assessment design audit

1. Define the construct. What capability is the assessment trying to interpret?
2. Define unsupported performance. What conditions apply before help is introduced?
3. Specify the assistance ladder. What information does each level add?
4. Define the stopping rule. When does prompting stop?
5. Record the response path. Which help changed what?
6. Use parallel evidence. Avoid making the posttest a memory test of the taught example.
7. Remove help. Check immediate independent performance.
8. Change conditions. Test transfer.
9. Return later. Test whether improvement survives time.
10. Inspect assessor effects. Was mediation comparable?
11. Preserve uncertainty. Do not turn one dynamic session into a fixed trait.
12. Connect to action. What teaching move changes because of the result?

25. The missing-node scan

The missing node may be dynamic assessment when an unsupported test score is repeatedly treated as a complete explanation; when two equally low-scoring learners respond very differently to the first useful hint; when language or unfamiliar format may be suppressing performance; when tutoring records only whether an answer became correct but not what kind of help unlocked it; or when an adaptive system keeps assigning easier questions without ever testing whether a small intervention changes the route.

26. Evidence and limits

Dynamic assessment is not one single test format. Research spans cognitive, language, reading and educational settings, with different forms of mediation and different outcome measures. That diversity is a strength and a reason to resist one-size-fits-all claims.

Studies can support validity for a particular interpretation in a particular population without validating every prompt ladder or every “learning potential” score. Age, domain, language background, assessor training, task design and the interval to the criterion all matter.

The core scientific discipline is simple: state exactly what was changed, exactly what changed in the learner’s performance, and how far the evidence is allowed to travel.

The return path

Return to the learner who failed the first equation and succeeded after one hint.

The failure was real. So was the responsiveness. Neither should erase the other.

A static assessment would have told us what happened before help. A dynamic assessment adds a second observation: how the learner’s performance moved when a specified form of help entered the system.

Dynamic assessment is most useful when help becomes evidence—not when help becomes a reason to stop asking whether the learner can perform independently later.

Research and further reading

• Swanson & Lussier — A Selective Synthesis of the Experimental Literature on Dynamic Assessment
• Wood, Biggs & Molnar — Dynamic Assessments of Word Reading Skills in Diverse School-Age Children: A Meta-Analysis
• European Journal of Psychology of Education — Dynamic testing and cognitive flexibility
• Language Testing in Asia — Group dynamic assessment in EFL

eduKateSG Learning Node Series · 0234 · Previous: 0233 — How Knowledge Space Theory Works · Explore the How X Works Hub.