A four-year-old knows how to build a stable tower, tell a detailed story about a family trip and count a small collection accurately. During a formal assessment with an unfamiliar adult, she says almost nothing.
What exactly has been measured?
Assessing young children is difficult because development is uneven, context-sensitive and expressed through language, play, movement, interaction and emerging academic skills. A measure can look professional and still capture the wrong construct, produce unstable scores, burden children or teachers excessively, work poorly for multilingual learners, or generate information nobody can use.
High-quality Pre-K assessment therefore begins with fit for purpose. The question is not, “Which assessment is best?” It is, “Best for what decision, for which children, in which setting, with what evidence of reliability, validity, fairness and practical use?”
MDRC’s September 2026 work on high-quality Pre-K assessment frames the problem across five interacting dimensions: content, psychometrics, experience, usefulness and scalability, with multilingual considerations running through the framework. That is a valuable contemporary signal because it refuses to reduce assessment quality to one statistical number. A measure can be psychometrically strong and unusable in ordinary classrooms. It can be pleasant for children and too imprecise for the decision being made. It can cover the right domain and work differently across language groups.
This article owns one narrow canonical job on eduKateSG: Pre-K child-assessment quality as a fit-for-purpose system connecting developmental construct coverage, psychometric evidence, child and educator experience, usefulness, multilingual fairness and scalability. It does not own school readiness broadly, universal language screening, dynamic assessment, clinical diagnosis, special-education eligibility or psychometric theory generally.
The useful question is not, “How can we measure young children more?” It is: what information do we genuinely need, what does the chosen measure actually support us in saying, and how will the information improve the child’s learning environment?
The 50-second answer
High-quality Pre-K assessment works when the measure fits the construct, the child, the decision and the setting.
- Start with the decision. Instructional planning, programme evaluation, screening and diagnosis require different evidence.
- Define the construct. What exactly are you trying to measure?
- Check developmental content coverage.
- Inspect reliability and validity evidence for the intended population and use.
- Consider language and cultural fairness.
- Protect the child’s experience.
- Consider educator workload and training.
- Use several evidence sources when one measure cannot carry the decision.
- Return results quickly enough to affect action.
- Do not make higher-stakes claims than the evidence supports.
The shortest principle is: assessment quality is not the sophistication of the tool; it is the quality of the inference the tool supports for the decision you actually need to make.
1. Begin with purpose before instrument
A classroom observation used to plan next week’s teaching has a different job from a population measure used to evaluate a public programme.
A developmental screener has a different job from a diagnostic assessment.
Choosing a tool before naming the decision invites misuse.
2. One measure should not be expected to do every job
An assessment that is efficient for large-scale programme evaluation may be too coarse for individual teaching decisions.
A rich teacher observation may be too subjective for comparing thousands of programmes.
Match tool to inference.
3. Define the construct precisely
“Language,” “maths” and “self-regulation” are broad labels.
Does the measure target expressive vocabulary, grammar, listening comprehension, number comparison, inhibitory control or classroom engagement?
Construct clarity prevents vague scores from carrying excessive meaning.
4. Content coverage matters
A mathematics measure that tests counting only cannot represent all early mathematical development.
A language measure focused on vocabulary may miss syntax or narrative.
Check whether the sampled content matches the domain claim.
5. Development is uneven across domains
A child can show advanced spatial reasoning and emerging number-symbol knowledge.
A single overall readiness score can hide this pattern.
Profiles can be more useful than one rank when instruction is the goal.
6. Development is uneven within a domain
A child may count objects reliably and struggle with cardinality or comparison.
Good assessment can reveal which relation is secure.
Instruction improves when the evidence is granular enough to act on.
7. Reliability asks whether measurement is sufficiently consistent
Scores fluctuate because of child state, item sampling, observer judgement and ordinary measurement error.
A reliable measure reduces irrelevant fluctuation.
Required reliability depends partly on the stakes of the decision.
8. High-stakes decisions require stronger evidence
Using one brief measure to alter placement or label a child demands more confidence than using it to choose tomorrow’s small-group activity.
The consequence of error matters.
Raise the evidence standard as stakes rise.
9. Validity is about the interpretation, not a property stamped on the test
A measure can have evidence supporting one use and not another.
A vocabulary assessment may be valid for estimating one language domain and invalid as a global measure of intelligence.
Ask whether evidence supports the intended inference.
10. Norms need a relevant comparison population
A score can be compared with a normative sample.
If that sample differs substantially in language, culture or age, interpretation becomes weaker.
Inspect who was included in development and validation.
11. Criterion-referenced information answers a different question
Instead of asking where a child ranks, a criterion measure asks which skills or behaviours are demonstrated.
This can be particularly useful for instructional planning.
Ranking and mastery are not interchangeable.
12. Direct assessment captures performance under specific conditions
A child responds to tasks administered by an adult under a set protocol.
This can standardise opportunity and create unfamiliarity.
Performance should be interpreted as what the child demonstrated in that context.
13. Observation captures behaviour in natural activity
Teachers can see language, play, persistence and social interaction during ordinary routines.
Observation gains ecological richness and risks observer bias and inconsistent opportunities.
Training and structured rubrics improve quality.
14. Teacher judgement is evidence and not infallible
Teachers know children across time and contexts.
Expectations, relationships and cultural assumptions can influence ratings.
Use clear behavioural anchors and moderation.
15. Parent and family report can add context
Children may show skills at home that are not yet visible in school.
Families can report language use, routines and development.
Questions need accessible language and culturally respectful framing.
16. Child state matters more in early assessment
Hunger, fatigue, illness, separation anxiety or unfamiliarity can strongly affect performance.
One poor session should not automatically become a stable trait.
Repeat or corroborate where the decision is important.
17. Rapport can affect performance
Young children often respond differently to familiar and unfamiliar adults.
Standardisation may require unfamiliar administration in some research contexts.
Classroom decisions can often use repeated evidence from familiar relationships.
18. Play can reveal capability and complicate standardisation
Children demonstrate planning, language, number and social reasoning through play.
Open play is difficult to score consistently because children encounter different opportunities.
Structured play tasks can combine natural engagement with clearer observation targets.
19. Multilingual children require construct clarity
A child may know the concept and lack the assessment-language word.
If the intended construct is conceptual knowledge, language load can distort the result.
If the intended construct is English vocabulary, language performance is part of the target.
20. Bilingual knowledge should not be mistaken for two monolingual deficits
Words and concepts can be distributed across languages.
Assessment should acknowledge the linguistic repertoire where the decision allows.
Use qualified multilingual assessment approaches rather than improvised translation.
21. Translation can change item difficulty
A directly translated word may be more or less common in another language.
Sentence structure and cultural familiarity can change too.
Validated adaptations are stronger than last-minute translation.
22. Interpreter use needs appropriate protocol
An interpreter can support communication and can unintentionally cue answers or alter standardisation.
High-stakes assessment requires trained procedures.
Classroom observation can use more flexible multilingual evidence.
23. Cultural familiarity can affect task performance
A story, object or scenario can be more familiar to some children than others.
This may matter especially when background knowledge is not the construct being measured.
Item review should include diverse perspectives.
24. Disability access needs planned accommodations
Motor, hearing, vision, communication and attention differences can prevent a child from showing the target skill through the standard response mode.
Accommodations should preserve the construct where possible.
Changing the response mode can sometimes change what is being measured.
25. Experience quality is an assessment dimension
A technically strong assessment can be distressing, exhausting or developmentally inappropriate.
Young children deserve humane measurement conditions.
Child experience also affects data quality.
26. Session length matters
Long batteries can produce fatigue and declining attention.
Segment assessment where protocols allow.
Collect only information that serves a real decision.
27. Repeated testing can consume learning time
Assessment has opportunity cost.
More data are not automatically better data.
Every assessment system should justify its time burden.
28. Educator burden is part of quality
A measure requiring hours of scoring per child may not scale reliably across ordinary programmes.
Teachers may rush or abandon protocols.
Feasibility affects fidelity.
29. Training requirements should match complexity
Some tools can be used accurately after brief training. Others need extensive certification.
Do not deploy a complex measure without the expertise it assumes.
Scalability includes human capacity.
30. Scoring should produce interpretable information
A percentile or composite can look authoritative and offer little guidance.
When teaching is the purpose, results should connect to observable capabilities and next learning.
Teachers need meaning, not only numbers.
31. Feedback latency matters
Results returned three months later may be useful for programme evaluation and useless for current classroom planning.
Decision timing should shape assessment design.
Useful evidence arrives before the decision closes.
32. Dashboards can create false precision
Colour-coded scores can imply sharp categories where measurement uncertainty is substantial.
Display confidence and context where possible.
Visual polish should not erase uncertainty.
33. Cut scores create consequences around small differences
A child just below a threshold can receive a different label from one just above.
Use multiple evidence sources when consequences are high.
Measurement error is especially important near cut points.
34. Screening should lead to a next step
Identifying risk without available follow-up creates anxiety and little educational value.
Before screening at scale, know how positive screens will be assessed or supported.
Capacity is part of ethical measurement.
35. Screening is not diagnosis
Screeners are designed to identify who may need closer assessment.
They prioritise sensitivity and efficiency differently from diagnosis.
Do not tell families a screener has established a clinical condition.
36. Programme evaluation requires different sampling
A public agency may need population-level trends rather than detailed individual profiles.
Shorter standard measures can be appropriate.
Do not force programme-evaluation instruments into classroom instructional use automatically.
37. Classroom assessment can be frequent and low stakes
Teachers observe counting, conversation, drawing and play daily.
Short evidence can guide grouping and activity design.
Not every useful observation needs a standard score.
38. Documentation should not crowd out interaction
An educator staring at a tablet to record every behaviour can miss the child.
Use efficient sampling and later note-taking where appropriate.
Assessment should serve the relationship, not dominate it.
39. Several weak measures do not automatically create strong evidence
Triangulation helps when sources provide partly independent information.
Three versions of the same biased observation do not solve the problem.
Choose complementary evidence.
40. Disagreement among measures can be informative
A child performs poorly in direct assessment and strongly during classroom observation.
Investigate context, language, rapport and task demands.
Do not average away the discrepancy before understanding it.
41. Assessment should not become a curriculum ceiling
Teachers can overfocus on tested indicators and neglect rich language, play, creativity or knowledge.
A measure samples development; it does not define all valuable learning.
Curriculum should remain broader than the instrument.
42. Good measures can still be misused for accountability
A child-development tool designed for formative use may become high-stakes when attached to staff evaluation or funding.
That can change behaviour around testing.
Use instruments only for purposes supported by evidence.
43. Data governance matters
Young children cannot meaningfully control long-term data systems.
Collect only necessary information, protect privacy and define retention and access.
Assessment infrastructure is also data infrastructure.
44. Families need understandable communication
“Standard score 87” means little without context.
Explain what was observed, uncertainty, strengths, next steps and what the result does not mean.
Do not turn a developmental snapshot into a permanent identity.
45. Strengths belong in the assessment conversation
Risk-focused systems can make every report a deficit list.
Record what the child can do and the conditions that support success.
Instruction builds from strengths as well as gaps.
46. Assessment systems should be reviewed like curricula
Measures age, evidence improves and populations change.
Revisit validity, language access, burden and usefulness periodically.
Do not keep an instrument because the data series is convenient.
47. Scalability can change quality
A measure administered carefully by researchers may perform differently when thousands of busy practitioners use it.
Implementation studies matter.
Quality at scale is not guaranteed by pilot evidence.
48. The endpoint is better decisions for children
An assessment is successful when it improves the accuracy and timing of decisions without imposing disproportionate burden or unfairness.
The score is not the product. The improved decision is.
Worked case 1 — The quiet multilingual child
A multilingual child produces few English words during a direct language task and tells elaborate stories at home in another language.
The teacher combines direct English-language evidence with family report, classroom observation and multilingual support before interpreting the profile.
The result becomes a language-development plan rather than a global cognitive label.
Worked case 2 — The screen without a pathway
A district introduces a broad developmental screener and identifies hundreds of children for follow-up. Specialist capacity was not expanded.
Families wait months with no clear next step.
The district changes rollout so screening capacity and follow-up assessment capacity grow together.
Worked case 3 — The beautiful dashboard
A Pre-K programme receives red, amber and green scores for every domain. Teachers begin treating children one point below a cut score as fundamentally different from children one point above.
Leaders retrain staff on measurement error, use continuous scores and combine evidence before high-stakes decisions.
The visual system becomes less categorical and more honest.
Worked case 4 — Observation burden
Teachers are asked to upload dozens of evidence notes and photos for every child each week. Documentation time expands while adult–child interaction falls.
The programme samples fewer high-value indicators and creates scheduled observation windows.
Assessment becomes more sustainable and learning time returns.
Practical route for early-years educators
Name the instructional decision first. Use observations and tasks that reveal the relevant construct. Record enough evidence to see patterns without documenting every moment. When direct measures disagree with daily behaviour, investigate context rather than forcing one answer.
Communicate strengths, gaps and uncertainty to families clearly.
Practical route for leaders and researchers
Review measures across content, psychometrics, experience, usefulness and scalability. Inspect validation populations and multilingual evidence. Match training and administration burden to real capacity. Do not use one instrument for purposes beyond its evidence.
Practical route for parents and families
Ask what the assessment was designed to measure, how your child experienced it, what other evidence is considered and what decision will follow. Share skills seen at home, including in other languages.
Treat one score as information, not destiny.
Common failure modes
- Choosing a measure before defining the decision.
- Expecting one assessment to serve screening, diagnosis, instruction and programme evaluation.
- Using broad labels without construct definition.
- Ignoring uneven development across domains.
- Using one brief session for a high-stakes conclusion.
- Confusing reliability with validity.
- Using norms from an irrelevant population without caveat.
- Treating teacher judgement as either perfect or worthless.
- Ignoring family and cross-setting evidence.
- Measuring multilingual children as if they were monolingual in two languages.
- Improvising translations of standardised items.
- Ignoring cultural familiarity.
- Failing to provide disability accommodations.
- Using developmentally distressing assessment procedures unnecessarily.
- Running batteries too long for young children.
- Creating documentation workloads that reduce teaching quality.
- Returning data too late to affect decisions.
- Treating cut scores as exact natural boundaries.
- Screening without follow-up capacity.
- Letting the assessment define the entire curriculum.
Frequently asked questions
What makes a Pre-K assessment high quality?
Fit among purpose, developmental content, psychometric evidence, child and educator experience, usefulness, fairness and the ability to implement the measure reliably at scale.
What are MDRC’s five dimensions?
Content, psychometrics, experience, usefulness and scalability, with multilingual considerations integrated across the framework.
Is direct testing better than observation?
Neither is universally better. Direct measures standardise tasks; observations capture natural performance. The best choice depends on the construct and decision.
Can one score show school readiness?
Young children develop unevenly across domains. A single composite can hide meaningful strengths and gaps and should be interpreted cautiously.
How should multilingual children be assessed?
Use measures and interpretations that distinguish language proficiency from other constructs, consider the child’s full linguistic repertoire where appropriate, and use validated multilingual methods when stakes are high.
What is the difference between screening and diagnosis?
Screening identifies who may need closer assessment. Diagnosis requires more specialised evidence and qualified interpretation.
How much assessment is enough?
Enough to support the decision reliably, but not so much that testing and documentation consume disproportionate learning time.
What should families receive?
Clear explanation of what was assessed, what the child demonstrated, uncertainty, strengths, next steps and what the result does not establish.
Can assessment data be used to evaluate teachers?
Only when the measure has evidence supporting that use. Instruments designed for child-level formative decisions may be invalid for high-stakes staff accountability.
What is the strongest success condition?
The assessment leads to a better, fairer and more timely decision for the child or programme without creating excessive burden or misleading certainty.
Evidence boundary and caveats
Pre-K assessment evidence is highly measure-specific. Psychometric properties depend on population, administration, language and use. Young children’s performance can vary more by context than older students’ performance, and standardisation can trade ecological realism for comparability.
The MDRC framework is useful because it treats technical quality, user experience and scale as interacting dimensions. It is a decision framework, not a claim that one assessment model is universally superior.
Sources and further reading
- MDRC, September 2026 high-quality Pre-K assessment toolkit and framework: https://www.mdrc.org/
- Education Endowment Foundation, early-years measures and assessment evidence resources: https://educationendowmentfoundation.org.uk/early-years/
Continue exploring on eduKateSG
- How X Works Hub
- How School Readiness Works
- How Checks for Understanding Work
- How Early Mark-Making Becomes Writing
- How Early Mathematical Comparison Works
- How Language Expansion Works
The final idea
Young children are not data points waiting to be extracted.
They are developing people whose knowledge appears differently across language, play, relationship and context. Assessment has to be technically strong enough to support a real inference and humane enough that the conditions of measurement do not overwhelm the thing being measured.
The best assessment system does not maximise the number of scores. It helps adults make better decisions while remaining honest about what the evidence can and cannot say.