Reader gateway: Imagine two education systems. In one, almost everybody says they enjoy solving unfamiliar problems. In the other, enthusiasm is rarer, but the pupils who express it tend to score much better than their classmates. Which country has ‘more effective curiosity’? The question sounds sensible until we notice it combines two different quantities. Average enthusiasm describes how common an attitude is. The relationship between enthusiasm and scores describes how two variables travel together among pupils. One can be high while the other is modest. A policymaker who fails to separate them may celebrate the wrong statistic or assign a school the wrong problem to solve.
Short answer: An education system can report a high average level of curiosity or perseverance without showing an especially strong relationship between that attitude and science scores. Conversely, a system with lower average levels may have a stronger within-system relationship. OECD’s PISA 2025 results explicitly warn that average curiosity levels are not necessarily highest in the systems where curiosity has the strongest association with science performance, and report a similar distinction for perseverance. The average, the spread, the relationship with scores, the distribution of prior achievement, measurement properties and the school environment are different pieces of the picture. None is a controlled experiment. For practical decisions, ask which quantity a chart shows, whether it compares students or countries, which factors are accounted for, and what a local teacher can genuinely change.
The 50-second route
Imagine Country A reports high average curiosity, but nearly every pupil reports similarly high curiosity. Because responses are bunched together, differences in curiosity may distinguish pupils’ scores only weakly. Country B reports lower average curiosity with greater variation across pupils, and within that system pupils with higher reported curiosity may show markedly different science performance. Both patterns are possible; neither proves that curiosity is more useful in one place. Averages answer ‘how much is reported on average?’ Associations answer ‘do students with different reported levels tend to have different scores?’ A correlation or regression coefficient also depends on measurement, variation and other conditions, not just the underlying learning mechanism. PISA 2025’s country comparisons illustrate why both quantities must be read separately. Do not infer from a national mean that an individual child is curious, or from a correlation that teaching curiosity will cause a particular score increase.
1. Two questions hidden inside one headline
Public conversation tends to compress evidence into a single league table. A country has an average mathematics score. A school has an average wellbeing result. A cohort has an average curiosity response. The number seems to summarise a large institution, which makes it attractive for charts and headlines. But educational attitudes and achievement are not simply one ladder. We might want to know the prevalence of an experience, the diversity of experiences, the fairness of access to it or its relationship with other outcomes. These are not the same inquiry.
Take curiosity as an example. Asking ‘How curious are students here?’ is primarily about distribution. If a questionnaire is well designed for the setting, the reported responses can show whether many pupils express interest in learning. Asking ‘Does curiosity distinguish higher and lower science performance?’ is about an association between two quantities. It requires paired observations and attention to how the association is estimated. A system can succeed in making curiosity widely reported and still have other factors such as knowledge, language, instruction and assessment strongly shaping performance differences.
Teachers also encounter this distinction. Two classes might have equally positive average attitudes yet radically different relationships between confidence and errors. One class contains confident students with both high and low attainment. Another contains a clear confidence-performance gradient. A blanket instruction to ‘raise confidence’ ignores the actual problem in each class. The question must be chosen before the indicator.
2. The international evidence that prompted this article
The OECD published PISA 2025 Results, Volume I, on 8 September 2026. Its chapter on student attitudes towards learning reports positive associations between curiosity and science performance, after accounting for students’ and schools’ socioeconomic profiles, across participating systems with available data. Yet it states that the countries with the highest average curiosity are not necessarily those where curiosity is most strongly associated with science performance. It makes a comparable observation for perseverance. This is a useful analytic lesson precisely because an apparently simple ‘more is better’ narrative would miss it.
The report’s result has limits. The participating students are 15-year-olds in the assessed systems, not a randomised intervention group. Responses about attitudes are self-reported and may reflect interpretation, language and social norms. Performance is measured through PISA’s assessment framework; it is not a direct readout of every valuable educational outcome. Statistical adjustment reduces some measured differences, but it cannot eliminate all confounding or reveal a unique causal direction. The article therefore treats the report as a reason to read indicators carefully, not as a formula for engineering school motivation.
To understand the finding, we can separate four questions: where the distribution is centred, how widely it spreads, how the two variables associate and what the measurements allow us to compare. These questions belong beside one another on the desk. They should not be collapsed into one claim that Country A ‘uses curiosity better’ than Country B.
3. The mean: a useful number with a limited job
An average indicates the centre of observed or modelled values under a specified calculation. Suppose a fictional survey codes four ordered response categories numerically and reports an average. The figure may be a convenient summary, but its interpretation depends on the response scale and analytic model. An ordinal agreement choice is not automatically a physical ruler with equal distances between every category. Major international assessments often use scaled indices built from multiple items. A reader should consult the technical definition rather than assume the number is the mean of a single question.
Even after a valid summary has been constructed, a mean hides distribution. Two classes can have the same average with different patterns: nearly everyone clustered near the middle, or some highly positive and others strongly negative. These patterns can call for different support. If half a class feels excluded while the other half feels highly connected, the average may appear acceptable while concealing an urgent equity problem. Conversely, a high average may mean an experience is broadly shared, which can be educationally valuable even if it no longer distinguishes examination performance.
Averages should therefore be presented with context: who responded, what items were asked, how scores were scaled, what uncertainty exists and how responses vary. A number without these details invites invented stories. The appropriate first question for a reader is not ‘Is the mean impressive?’ but ‘Which experience does this number summarise, for which pupils, under which measurement rules?’
4. The slope: an association, not a second way to say mean
Now imagine plotting each pupil’s reported curiosity against their science score. A relationship measure asks whether the two variables tend to change together. In a simple linear model, a slope summarises the expected difference in the outcome associated with a specified difference in the predictor, given the model’s assumptions. A correlation instead expresses the strength and direction of a standardised linear association. An R-squared measure summarises a model’s account of outcome variation under its definitions. These related quantities are not interchangeable and should not be described casually as an ‘impact’.
Two systems can have identical mean curiosity but different slopes because students’ experiences and performance differ in how they co-vary. They can have different means and similar slopes. One may have a narrow range of attitude scores, limiting the observable differentiation between pupils. Another may have larger spread or more measurement noise. Prior knowledge, school composition, instructional quality and learning opportunities may shape the association. None of these possibilities by itself proves a mechanism. They show why the mean does not mathematically determine the relationship.
This matters in school improvement conversations. If a school celebrates a rise in average self-reported motivation, leaders should not announce that motivation now explains more of its examination results. The second claim requires a separate analysis. Likewise, a stronger motivation-score correlation does not establish that overall motivation has improved. It could reflect growing differences among pupils, which may be an equity warning rather than a success.
5. A fictional two-country example with a clear warning label
Consider two invented systems, Harbour and Ridge. In Harbour, most pupils report moderately high curiosity. Their science scores range widely because prior knowledge, reading comprehension, instructional exposure and other factors vary substantially. In Ridge, average reported curiosity is lower, yet pupils differ more in how often they investigate unfamiliar ideas. In the fictional dataset, the higher-curiosity pupils also tend to have higher science scores, producing a visibly steeper relationship.
The conclusion is not that Ridge has discovered a superior motivation programme. Perhaps Ridge’s curiosity scale is functioning differently. Perhaps Harbour’s nearly universal positive answers compress useful variation. Perhaps differences in access to enrichment opportunities cause both curiosity and scores to rise among particular students in Ridge. Perhaps the stronger relationship changes after accounting for family background and prior achievement. Perhaps the data contain fewer observations at one end of the scale. Without further evidence, ‘stronger’ describes the statistical relationship, not the causal power of a trait.
The example is deliberately simplified. Actual PISA comparisons use sophisticated sampling, weighting, plausible values, uncertainty estimation and reporting conventions. An illustrative scatterplot in a school newsletter is not a substitute for that methodology. But the conceptual lesson transfers: two separate questions require two separate indicators, and neither one earns an explanation merely by looking impressive.
6. Why range restriction changes what a relationship can show
Imagine a class where almost every student reports that science is important. If the measure is near its ceiling for many pupils, observed responses contain little variation. A correlation with science scores may be small partly because the questionnaire has little room to distinguish students who all choose the upper category. That does not mean the attitude is unimportant. It means the instrument may not discriminate the intended construct well in that particular range.
The opposite problem can also occur. A short survey with vague items may produce noisy differences between students. Measurement error can weaken observed relationships. If one student interprets ‘I enjoy learning new things’ as love of laboratory investigations and another interprets it as enjoyment of social media tutorials, the same response may summarise different experiences. Technical psychometric work is needed to evaluate such problems, especially across languages and cultures. eduKateSG already has a separate canonical guide to measurement invariance; this article uses that as a caveat rather than attempting to replace it.
Teachers should resist the impulse to fix a low correlation by redesigning their questionnaire until it produces a stronger one. A measure should represent the intended construct, not be engineered to create a more persuasive story. The proper response to a ceiling effect may be additional, carefully validated indicators or more informative classroom evidence, not manipulation of the scale.
7. Why a common cause can create a relationship
Suppose pupils with strong prior reading skills understand science questions more readily and also enjoy exploring unfamiliar science material. Their prior reading capability may influence both reported curiosity and science performance. If we observe an association between curiosity and scores, part of it may reflect this common background. The same logic can apply to family resources, access to science experiences, teacher expectations, language proficiency or earlier success.
This is confounding in ordinary language: a third factor contributes to both quantities we are relating. Statistical adjustment for measured socioeconomic background can help, but the chosen variables, their measurement and their positions in the causal story matter. Conditioning on the wrong variable can sometimes introduce new distortions. No convenient ‘adjusted’ label guarantees causal identification. Researchers use substantive theory, designs and sensitivity checks to decide which interpretations remain plausible.
The classroom implication is modest but important. If curious pupils score higher, a school should not simply demand more visible curiosity and expect marks to rise. It should investigate whether pupils have the knowledge, access, feedback and instructional opportunities that make curiosity productive. A learner who is invited to investigate but cannot read the source material needs different help from one who understands the basics but rarely gets intellectually demanding questions.
8. Why the direction may also run backwards
Achievement can influence curiosity. A pupil who repeatedly understands explanations may become more willing to ask questions. A pupil who experiences confusion without adequate support may begin avoiding a topic they once enjoyed. Conversely, curiosity can encourage more reading, practice, inquiry and participation, which may build competence. Both directions are plausible, and both may operate in loops over time.
A single survey and assessment cannot simply untangle that sequence. Even if responses and test results are collected in a fixed order, prior experiences shape both. Longitudinal evidence can add information about change but still requires careful interpretation. Randomised tests of specific instructional changes can identify some causal effects under particular conditions, but they do not make the abstract trait ‘curiosity’ a single switch that can be toggled for every child.
Think of the learner who initially enjoys chemistry but lacks basic proportional reasoning. Repeated unsuccessful calculations may reduce willingness to investigate. Teaching the mathematical prerequisite may improve performance and preserve interest without an explicit ‘curiosity programme’. Another pupil may know the chemistry content but find every task repetitive; a better question can increase engagement without changing the core explanation. Different starting points require different hypotheses.
9. Between countries and within countries: never switch the unit silently
A country is not a student enlarged to national scale. Between-country comparisons involve national averages and systems. Within-country comparisons involve differences among individual pupils or schools inside a system. The two levels can have different relationships because the sources of variation are different. School structures, cultural practices, economic conditions and assessment arrangements influence national patterns, while classroom experience, prior knowledge and family circumstances influence individual variation.
An ecological error occurs when someone takes a group-level relationship and assumes that it describes each individual. A related mistake appears when a student-level association is turned into a claim about countries. A system where curious pupils tend to score higher does not necessarily have a higher average science score than a less curious system. Conversely, a high-performing system may contain many pupils who report limited enthusiasm for school science.
EduKateSG has broader canonical material on aggregation-level reasoning. The particular contribution here is the PISA 2025 mean-versus-within-system-association interpretation. That narrow boundary matters because it prevents this new article from becoming an unnecessary duplicate of general statistical or measurement lessons.
10. The school-level version: two classes with equal averages
At a fictional Singapore secondary school, Classes A and B both report similar average confidence in mathematics. In Class A, nearly everyone is moderately confident; independent problem-solving performance still varies widely. In Class B, some pupils are very confident and others strongly doubtful, with the more confident group also showing higher marks. A simple headline says both classes are equally confident. A more detailed analysis asks whether confidence differentiates attainment in either class, whether the scale is sufficiently sensitive and what the distributions conceal.
The next action should not be to copy the same confidence intervention into both rooms. In Class A, teacher observations may reveal that pupils overestimate understanding after seeing worked examples. The need is better independent checks. In Class B, the lower-confidence group may lack prerequisites or face repeated public embarrassment. The need may be targeted teaching and safer participation. Even then, these diagnoses require work samples and conversations; they cannot be recovered solely from the survey graph.
The lesson is not that classroom surveys are useless. It is that they are starting points for enquiry. A mean can prompt a question about prevalence. A relationship can prompt a question about differentiation. Neither replaces subject-specific evidence about how pupils actually solve problems, read passages or explain mechanisms.
11. The Primary 6 example: parents comparing averages across tuition groups
Two tuition groups report that their pupils enjoy mathematics. Group One has uniformly positive responses but a wide range of examination readiness. Group Two has a lower average enjoyment score and a sharper link between enjoyment and recent marks. A parent could be tempted to conclude that Group Two has found a teaching method that turns enjoyment into results more efficiently. That conclusion is not supported by these summaries.
The pupils may have entered with different prior attainment. One group may include children who enjoy number games but still struggle to interpret word problems. Another may contain a concentrated group with both strong foundations and positive experiences of success. Class size, attendance, curriculum exposure, question difficulty and selection into tuition can influence the figures. If the surveys use different wording, even the averages may not be directly comparable.
A better parent question is concrete: what can my child do independently now that they could not do before? Ask for a fresh problem, an explanation of the chosen method and a later retrieval check. Interest matters, but a credible learning plan links it to the actual capability needed. Choosing a tutor should not become an attempt to optimise a poorly defined correlation.
12. The English example: high enthusiasm does not eliminate a skill bottleneck
A class enjoys discussing novels. Students speak warmly about stories, characters and film adaptations. Yet some struggle to support an inference with specific textual evidence. The average enthusiasm is high; the relationship between reported enthusiasm and essay quality may be weak because other skills create the main difference. A teacher who treats the weak relationship as evidence that interest is pointless has misunderstood the finding.
Interest may help pupils keep reading, notice patterns and invest time. But essay performance also requires interpreting a question, selecting evidence, making a warranted claim, organising a paragraph and controlling language. If those competencies vary independently of enthusiasm, they can limit the score relationship. The instructional response is to teach the bottleneck while using interest as an access route.
For example, ask all pupils to explain why two readers might infer different motives from the same sentence, then require each interpretation to cite the language that supports it. The task connects engagement to a disciplined academic move. Later, test that move on an unfamiliar passage. You are no longer asking whether enthusiasm alone predicts marks; you are teaching a process that turns attention into justified interpretation.
13. The science example: curiosity needs an evidence route
A pupil loves astronomical questions, asks whether life exists elsewhere and follows science videos. On a school assessment, the pupil struggles with graph interpretation and the distinction between a controlled variable and an outcome. The enthusiasm is genuine. So is the skill gap. Treating the two as incompatible would reduce science education to either affection or scores, when both matter in different ways.
A teacher can use the interest to supply a meaningful context while explicitly teaching the measurement skill. Suppose the class examines how an observed brightness changes over time. Before making an explanation, students identify axes, units, uncertainty and the kind of claim the data can support. They compare two proposed explanations and ask which additional evidence would distinguish them. The curiosity does not automatically do the analytic work, but it can make that work worth doing.
Similarly, perseverance can keep a pupil working without guaranteeing that the chosen method is effective. Persisting with a flawed procedure may reinforce the error. In PISA 2025’s reporting, average perseverance and the strength of its association with performance are not interchangeable either. Good instruction connects persistence to feedback, correction and purposeful strategy change.
14. What an improvement in the mean would actually mean
Suppose a school uses the same well-understood survey process over several terms and finds that average curiosity rises. With appropriate caution about response patterns, sample composition and measurement stability, the school may have evidence that students report a more curious learning experience. That could be a desirable outcome on its own. Education has purposes broader than maximising test scores, and students’ relationship with learning matters.
But the institution should not claim an attainment effect unless it has appropriate evidence about achievement. The rise may reflect an enjoyable curriculum change, better classroom climate, altered wording, different students responding or a temporary novelty effect. A local evaluation can pair survey results with task-specific evidence: do pupils now generate better questions, test alternatives more carefully or apply knowledge in unfamiliar situations?
The school should also ask whose experience improved. A higher average can conceal a widening gap if already-engaged pupils benefit much more than others. A consistent improvement across groups is a different result from a dramatic increase for a small subset. Once again, the distribution matters as much as the centre.
15. What a stronger association would actually mean
Suppose the relationship between reported curiosity and science scores becomes stronger in a later cohort. This may mean curiosity-related differences align more closely with attainment differences in the observed data. It does not automatically mean the school has become better at fostering curiosity or converting it into achievement. The change could reflect wider inequality in access to inquiry, a change in the ability distribution, greater reliability of the survey, different topics or unmeasured changes in the cohort.
A stronger relationship can sometimes be troubling. Imagine that only students in one enrichment pathway receive opportunities to pursue substantial science questions, and those students already have more resources and higher attainment. Curiosity and scores may become more strongly associated because opportunities have become more stratified. The coefficient would look impressive while the educational distribution became less fair.
A weaker relationship can also have multiple meanings. It might reflect measurement noise, reduced range, more equitable instruction that helps pupils regardless of initial attitude, or an actual weakening of the mechanism. Without a design and contextual evidence, there is no universal interpretation. An honest report states the possible meanings and does not automatically rank the larger coefficient as better.
16. A five-step reading protocol for parents and journalists
Step one: identify the quantity. Is the headline showing a mean, a percentage agreeing with an item, a correlation, a regression slope, an adjusted difference or a model’s explained variation? Each has a different job. If the article calls them all ‘impact’, be cautious.
Step two: identify the unit. Are the dots students, schools, countries or changes over time? Do not switch units when explaining the result. A national average is not evidence that a particular child has the average attitude.
Step three: identify the measurement. What questions produced the scale, how was it constructed and what does one unit mean? Do not compare raw coefficients across incompatible units. Ask whether the measure is comparable across the groups being discussed.
Step four: identify the comparison and uncertainty. Which factors were considered, what design was used, and how precise is the estimate? ‘Adjusted’ does not mean all alternative explanations have disappeared.
Step five: identify the practical claim. Is the writer recommending a classroom intervention? If so, which action is expected to change which outcome, and what evidence supports that causal link? A descriptive international chart can inspire a question without providing the answer.
17. Teachers: use indicators to select hypotheses, not children
A useful survey should help teachers ask better instructional questions. If pupils report high value but low curiosity, inspect whether lessons include accessible investigation. If pupils report high curiosity but struggle on particular tasks, inspect prerequisite knowledge, reading load, representation and feedback. If curiosity is strongly related to scores, ask whether access to deeper tasks is uneven. These are hypotheses to test with work, not labels to attach permanently to pupils.
Student-level decisions need local, specific evidence. A child who says they dislike science may still reason accurately about fair tests. Another who loves science may need help distinguishing observation from explanation. A child who writes little in a survey may be tired, uncertain about wording or cautious about privacy. A teacher should not infer a complete motivational profile from a single response.
Pair each hypothesis with a low-cost check. Provide a novel question, ask for an explanation, use a short error analysis, or compare two versions of a task with different scaffolding. Keep the result proportionate: one lesson can justify a revised next lesson, not a universal claim about national motivation.
18. School leaders: build a dashboard that cannot tell an accidental story
A responsible dashboard should retain the meaning of its indicators. Present the average of an attitude index separately from any association with performance. Include a plain-language definition, the population, dates, response rate and uncertainty. Where privacy and sample size permit, show distribution or subgroup differences rather than only a total. Explain that student-level association and between-school comparison are separate analyses.
Do not ask the dashboard to assign motives to individual teachers or pupils. A teacher with a class showing a weaker attitude-score association may have produced more equitable learning, inherited a different distribution or used a survey with poor discrimination. The number requires interpretation. Likewise, a high class average may reflect selection or response style, not simply excellent pedagogy.
Leadership can encourage an evidence culture by asking for competing explanations in every report. What else could create the pattern? Which additional observation would distinguish those explanations? Which change is actually within the school’s control? A dashboard becomes useful when it supports decisions under uncertainty rather than rewarding the most flattering narrative.
19. A local evaluation route that respects the distinction
Choose a concrete instructional question rather than trying to raise an abstract national index. For example: can a short, well-scaffolded prediction-and-explanation routine improve pupils’ ability to interpret unfamiliar science data? Define the target performance and the comparison condition. Keep teaching time and task difficulty as comparable as practical. Gather baseline work, implement the routine and assess on fresh items that do not merely repeat the lesson.
If you also ask pupils about curiosity, treat it as a separate outcome or mechanism indicator with clear limits. The intervention may improve explanations without increasing reported enjoyment. It may increase curiosity before a detectable achievement change. It may benefit some pupils more than others. These patterns are not contradictions; they tell you which part of the proposed mechanism needs more evidence.
A small classroom trial will rarely support broad causal claims. It can still improve local practice if the team records changes honestly, avoids selecting only positive outcomes and checks whether the benefit survives a new context or a later date. The statistical lesson and the teaching lesson converge: ask a question your evidence can answer.
20. Failure modes: the shortcuts that produce misleading conclusions
Ranking the wrong quantity. A high mean and a strong association are different claims. Label every graph clearly before interpreting it.
Calling a coefficient a causal impact. An observational relationship can reflect common causes, reverse direction or measurement effects. Reserve causal language for evidence and assumptions that support it.
Switching from countries to children. A system average does not identify an individual’s attitude. Use individual evidence for individual support.
Treating a weaker correlation as a weaker school. Range restriction, reliable equitable teaching or different variation may weaken a relationship without reducing educational quality.
Assuming more variation is good. A stronger relationship can emerge because opportunities and achievement have become more unequal. Inspect distributions.
Using a poor survey to justify a precise action. If the measurement is unclear, fix the evidence before acting as though the resulting decimal is exact knowledge.
Ignoring what happens in the lesson. Neither means nor slopes explain how a pupil selects a method or evaluates evidence. Return to work samples and teaching mechanisms.
21. Why the lesson matters beyond curiosity
The same distinction appears in confidence, belonging, perseverance, wellbeing, perceived teacher support and attitudes toward digital learning. A school can have high average support yet little difference in scores associated with support because most students experience it. Another can show a strong relationship because access is uneven. Averages tell us something about the educational environment; relationships tell us something about observed differentiation. Both are important, but neither should swallow the other.
For parents, the best question remains practical: what does my child’s work reveal, and which support is appropriate now? For teachers, it is instructional: which mechanism will change the learner’s next decision? For leaders, it is systemic: are opportunities and outcomes improving fairly, and what do our measures actually allow us to conclude? PISA 2025’s mean-and-association puzzle is valuable because it interrupts the habit of turning every positive attitude into one competitive league table.
Education improves when evidence becomes more discriminating, not merely more abundant. A school that knows what its numbers mean can act with greater confidence and less exaggeration. A family that recognises the limits of a national chart can focus on the learner in front of it. The goal is neither statistical cynicism nor enthusiasm for every measurement. It is a clear match between the question, the unit, the evidence and the decision.
Frequently asked questions
Does a high average curiosity score mean curiosity strongly predicts achievement?
No. The mean describes central tendency. A relationship with achievement describes how differences among pupils co-vary. These can move independently.
Does a stronger curiosity-score relationship mean curiosity teaching works better there?
No. A relationship can reflect selection, prior knowledge, opportunity, measurement and other conditions. A causal claim about teaching requires evidence on a specified intervention.
Why might a very curious class show a weak correlation?
If nearly everyone reports high curiosity, the measure may contain little variation. Differences in scores may be driven mainly by other factors. Measurement error or task differences can also matter.
Is a correlation the same as a regression slope?
No. A correlation is a standardised measure of linear association. A regression slope expresses an outcome difference per specified predictor unit under a model. They are related but have different interpretations.
What is the ecological fallacy?
It is an error of applying a group-level relationship to individuals without justification. A country’s average cannot tell you the attitude or performance of a specific pupil.
Why does socioeconomic adjustment matter?
Background factors can be related to both attitudes and scores. Adjustment can help describe associations after accounting for measured variables, but it does not remove every alternative explanation or automatically identify causality.
Can a weaker association be good news?
Possibly. If effective teaching helps pupils across initial attitude levels, the score difference associated with that attitude might shrink. But measurement and composition changes could also explain it. Investigate before celebrating.
Should a school measure curiosity at all?
It can, if it has a clear purpose, suitable instruments, privacy protection and a plan for interpretation. Surveys should complement, not replace, evidence of actual reasoning and learning.
Should parents choose a school with the highest curiosity mean?
No single indicator can settle school fit or quality. Ask what the measure means and consider curriculum, teaching, support, safety, access, learner needs and evidence of actual capability.
Can we compare percentages from two differently worded surveys?
Not casually. Wording, response categories, sample and administration can change results. Apparent differences may partly reflect measurement rather than real changes in the underlying experience.
Does PISA 2025 prove every system has the same mechanism?
No. It reports patterns and associations with specified data and methods. The mechanisms producing those patterns can differ and require further investigation.
What is the first question to ask when seeing a headline statistic?
Ask exactly what the number measures and at which level: individual pupil, school, country or change over time. Many misleading claims become visible as soon as those are separated.
Surgical internal links
For the broader statistical tools, see How Ecological Fallacy Works, How Aggregation Bias Works and How Sampling Distributions Work: Center and Spread. The separate psychometric owner is How Measurement Invariance Works. For teaching context, return to How Education Works, the How X Works hub and the Clementi Secondary 1 Mathematics reference.
Sources and reading cautions
- OECD (8 September 2026), PISA 2025 Results (Volume I): Student attitudes towards learning. See especially the discussion of average curiosity and perseverance compared with the strengths of their associations with science performance.
- OECD, PISA 2025 Results Volume I. Report metadata and methodology references.
- Education Endowment Foundation, Metacognition and Self-regulated Learning guidance. Classroom strategy background; not evidence that the national correlation patterns are causal.
- Institute of Education Sciences / What Works Clearinghouse, Organizing Instruction and Study to Improve Student Learning. Evidence-informed classroom methods, separate from aggregate PISA associations.
Harbour, Ridge and all pupils described in cases are invented. The article makes no claims about unreported Singapore coefficients or the relative causal effectiveness of different nations’ curiosity. Any educational use of survey comparisons should consult the original technical documentation and limitations.