How Education Works | School Improvement — How Evidence, Implementation and Continuous Learning Change a School

Evidence boundary: all school cases, timetables and numerical examples below are invented teaching scenarios. They are not performance claims about eduKate or any real school. External sources are used for implementation and continuous-improvement principles; local adoption still requires professional judgement, context and the rules of the institution.

Reading route: Define the problem · Build a theory of change · Implementation · Worked school case · Measurement and review · Sustain or stop · Sources.

1. Begin with a problem that can be observed

“Raise standards” is a direction, not a diagnosis. “Improve reading” is more specific but still broad. A usable improvement problem identifies where the current process breaks. For example: students can retrieve stated information from a text but often cannot justify an inference using evidence from two parts of the passage.

The sharper problem changes the response. A general reading campaign may not address inference. A targeted sequence of modelling, guided comparison and independent evidence selection might. The school should therefore define the observed gap before purchasing a solution.

The U.S. Institute of Education Sciences provides a continuous-improvement toolkit for schools and districts built around iterative inquiry, including Plan–Do–Study–Act cycles and tools for identifying causes and drivers. Source: IES, Continuous Improvement in Education.

At school level, define the problem using several kinds of evidence where feasible: student work, assessment patterns, classroom observation, learner experience and teacher knowledge. One source can suggest where to look; several complementary sources can reduce the chance that the school improves the wrong thing.

2. Distinguish symptoms, mechanisms and causes

Low homework completion is a symptom. Possible mechanisms include unclear instructions, excessive workload, weak prerequisite knowledge, limited access outside school, low perceived value or an inconsistent submission process. The school should not select a cause simply because it fits an existing belief.

Write competing explanations. Then ask what observation would make one more likely than another. If students complete short in-class versions successfully but do not submit the same work online at home, access or process deserves investigation. If they cannot begin the work even with support in school, the learning design may be the more immediate issue.

A cause map is useful only when it leads to discriminating questions. A large diagram filled with arrows can create an impression of sophistication while leaving the school no clearer about what to test next.

3. Choose a priority by educational importance and tractability

A school has more legitimate needs than it can address simultaneously. Prioritisation therefore requires trade-offs. Ask how strongly the problem affects worthwhile learning, how many learners it affects, whether the school has authority to change the relevant process, and what work will be displaced.

A problem can be important but poorly defined. Another can be easy to measure but educationally minor. The improvement agenda should not become a list of whatever generates the cleanest dashboard.

Limit concurrent major changes. If a school introduces a new curriculum, assessment platform, behaviour system and reporting structure at once, disappointing results become difficult to interpret. Teachers also lose the capacity to learn each change properly.

4. Build a theory of change before implementation

A theory of change explains how the proposed action is expected to produce the desired result. It can be written in plain language: “If teachers model how to select textual evidence, then give students guided and independent practice on increasingly unfamiliar passages, students should become better able to justify inferences without prompts.”

The statement exposes assumptions. Teachers need suitable examples. Students need enough vocabulary to access the passage. Independent practice must actually remove the crucial prompt. The assessment must test evidence selection rather than only asking students to identify a teacher-modelled answer.

For each assumption, identify what the school can observe. This turns the improvement design into a testable system. If teachers never receive the planned preparation, a disappointing student result should not be interpreted as evidence against a programme that was never meaningfully implemented.

5. Evidence-informed selection is different from evidence-proof selection

Research can help identify approaches with credible supporting evidence, but no external study removes the need to examine local fit. The learners, timetable, staff expertise, curriculum and implementation conditions may differ from the research setting.

Ask what the evidence supports, for whom, under what conditions and with what outcomes. Then compare those conditions with the local problem. A promising programme for early reading is not evidence for solving a secondary mathematics misconception merely because both involve “intervention.”

The Education Endowment Foundation’s implementation guidance encourages schools to make evidence-informed choices, prepare deliberately, deliver with support and sustain selectively. Its 2024 third edition organises implementation around behaviours, contextual factors and a structured but flexible process. Source: EEF, A School’s Guide to Implementation.

6. Define the active ingredients

An active ingredient is a feature believed to be necessary for the approach to work. In the reading example, active ingredients might include teacher modelling of evidence selection, guided comparison of strong and weak evidence, independent retrieval from unfamiliar passages and feedback that requires another attempt.

Other features may be adaptable: the passage topic, exact worksheet layout or order of some examples. Distinguishing the essential from the adaptable gives teachers room to use professional judgement without accidentally removing the mechanism.

Write the distinction before launch. Otherwise a programme can drift until every classroom is “doing the initiative” under the same name while the educational process differs substantially.

7. Implementation is a social and operational process

EEF’s 2024 guidance identifies engage, unite and reflect as behaviours that support implementation, alongside attention to contextual factors such as the approach itself, systems and structures, and people who enable change. Source: EEF, implementation behaviours.

For a school, that means teachers need more than an announcement. They need to understand the problem, the intended mechanism, what will change in their work, what can be adapted, where help is available and how the school will know whether implementation is progressing.

Participation should be meaningful without making every design decision a referendum. Leaders can set a clear direction while inviting teachers to identify practical obstacles and improve the route. People closest to classroom work often see dependencies that a central plan misses.

8. Prepare the infrastructure before declaring launch

A change may require timetable space, materials, professional learning, assessment items, moderation, data access or specialist support. If these are missing, the initiative begins with hidden debt. Teachers compensate individually until the variation becomes too large or workload becomes unsustainable.

Make an implementation readiness list: who leads; who supports; what training is needed; which resources must exist; what time is protected; what evidence will be collected; what happens when a teacher or student joins late; and which existing process is being replaced.

Replacement matters. Improvement cannot be an infinite accumulation of new practices. If a new feedback routine is adopted, identify which older marking expectation can be retired. If a new assessment system is introduced, stop collecting data that no longer serves a decision.

9. Train the decision, not merely the material

Professional learning should help teachers understand when and why to use the approach. Giving everyone a resource pack is not the same as developing implementation capability. Teachers need opportunities to see examples, rehearse decisions, compare responses and receive feedback.

In the reading example, teachers might jointly analyse two student responses and decide whether each needs more work on inference, evidence selection or explanation. This builds shared judgement around the intended mechanism.

Keep preparation proportionate. An improvement programme that consumes so much training time that ordinary teaching deteriorates has created another problem. The design should preserve the highest-leverage elements and make them usable.

10. Use early implementation evidence before outcome evidence is mature

In the first weeks, it may be too soon to detect stable changes in attainment. The school can still inspect whether the planned practice is appearing. Are teachers using the model? Are students attempting the independent step? Are the new materials available? Is feedback followed by another attempt?

These implementation indicators do not prove effectiveness. They answer a different question: did the intended process occur? If not, the next action may be to repair implementation rather than judge the educational theory prematurely.

Keep observations supportive and specific. “Teacher not implementing initiative” is a poor note. “Independent task retained the same highlighted evidence used in the model, so students did not need to select evidence themselves” identifies a design problem that can be corrected.

11. Worked case: improving evidence use in written explanations

Invented school case: teachers notice that many students quote information in English and humanities answers but do not explain how the evidence supports the claim. A review of 120 anonymised pieces of work finds that relevant quotation is more common than explicit reasoning between quotation and conclusion. The number 120 is invented for this example.

The school defines the problem narrowly: students can often locate relevant evidence but are inconsistent at articulating the inferential link. That diagnosis prevents a generic “write more” campaign.

The proposed mechanism has four parts. Teachers model a short claim–evidence–reasoning sequence. Students compare two examples where the evidence is identical but the reasoning differs. They then complete a partially guided explanation before attempting an independent passage. Feedback targets the missing logical link and requires a revision.

The school identifies active ingredients: comparison of strong and weak reasoning, a visible inferential step, independent evidence use and revision after feedback. Teachers can choose subject-specific texts and terminology while preserving these features.

Before launch, department leaders prepare examples, agree a small observation rubric and remove one older writing template that conflicts with the new approach. Staff practise scoring several student explanations so that “reasoning present” has a more consistent meaning.

During the first month, the school samples lessons and student work. It discovers that most teachers use the model but some independent tasks still contain sentence stems that effectively supply the reasoning. The immediate repair is to keep the stem for guided practice but remove it from the later task.

After a suitable interval, a new set of unfamiliar passages is used. The school compares the proportion of responses containing defensible evidence-to-claim reasoning, while also inspecting overall writing quality and whether the task difficulty changed. An improvement would be encouraging evidence, not automatic proof that the initiative alone caused the change.

Teacher workload is reviewed separately. If the approach requires extensive extra marking, the school explores whether feedback can be shorter and still lead to a meaningful revision. Sustainability is part of effectiveness because a practice that cannot be maintained will not remain part of the learning environment.

12. Use Plan–Do–Study–Act as a learning loop, not a ritual

The IES continuous-improvement toolkit introduces Plan–Do–Study–Act as one framework for iterative improvement. The value is not the acronym itself. It is the discipline of making a prediction, trying a bounded change, examining what happened and deciding what to do next. Source: IES toolkit.

Plan: state the change, prediction and evidence. Do: implement on a defined scale. Study: compare the observed process and outcomes with the prediction. Act: adopt, adapt, expand, pause or abandon according to the evidence.

A cycle should be long enough for the relevant mechanism to operate but short enough to support learning before the whole school has committed irreversibly. The appropriate interval depends on the change; there is no universal two-week or term-long requirement.

13. Pilot when uncertainty is high and the cost of error matters

A pilot can expose practical problems before full-scale adoption. It is particularly useful when the school is uncertain about timetable demands, training needs, data collection or student access. The pilot should be representative enough to reveal those issues.

Do not oversell pilot findings. A highly supported early group may perform differently from a later whole-school rollout. The people running the pilot may be unusually experienced or motivated. Treat the pilot as evidence about feasibility and initial promise, not as guaranteed proof of future impact.

Record what additional support the pilot received. If scale-up removes that support, the school has changed the intervention and should expect the result to change too.

14. Improvement data need a decision attached

Schools can collect enormous quantities of information with little improvement. Before adding a measure, ask: what decision will this inform; how often can that decision change; who will inspect the information; and what action follows different patterns?

A weekly dashboard is unnecessary for a decision reviewed once per term. A termly average is too slow for detecting whether a new classroom routine is not being used. Match the frequency of information to the speed of the mechanism.

For measurement principles, see Educational Measurement. An improvement system needs to know what its indicators actually support claiming.

15. Separate implementation, learning and side effects

A useful review has at least three columns. Implementation: did the intended practice occur? Learning: did learner capability change in the intended direction? Side effects and burden: what happened to time, workload, access, motivation or another important outcome?

Improvement can fail differently across the columns. Strong implementation with weak learning may challenge the theory or the chosen approach. Weak implementation with weak learning does not test the approach cleanly. Strong learning with unsustainable burden may require redesign before scale.

Keep the evidence traceable. If a headline says “writing improved,” preserve what changed in the task, scoring and population. Do not compare two percentages whose assessments or denominators differ substantially without stating that limitation.

16. Resist initiative bias

Schools can become biased towards visible additions: new programmes, platforms, slogans, forms and training. Yet the best improvement may be subtraction. Removing duplicated assessment, simplifying a transition or retiring an ineffective routine can return time to teaching.

Before approving a new initiative, ask what existing work it replaces. If the answer is “nothing,” calculate the burden honestly. Improvement should not depend on teachers silently extending their working day indefinitely.

Also ask whether the problem can be solved by improving the current system rather than adopting another. A weak help-seeking routine may need clearer instructions, not a new digital platform.

17. Leadership should protect the problem definition from drift

As a project grows, the original educational problem can disappear behind implementation activity. Meetings become about training attendance, resource completion and dashboard updates rather than the learner capability the project was supposed to improve.

Leaders should regularly restate the problem and mechanism. Ask whether each major task still serves them. If a data collection consumes time but no longer changes a decision, stop it. If a resource is popular but omits an active ingredient, repair it.

For the system-level role of leadership, see Educational Leadership. School improvement is where leadership claims become operationally testable.

18. Families and students can improve the diagnosis

When an initiative affects homework, assessment, transitions or access, students and families may observe parts of the process that school data do not show. Ask concrete questions: which instruction was difficult to interpret, what support was used, where did the process become impossible, and what changed after the new approach?

Consultation should add evidence, not transfer responsibility. A family can explain why a homework process is difficult to use; the school still owns the educational design it requires. A student can report that a routine is confusing; the teacher still decides how to preserve the learning purpose.

For a full framework, see Family–School Partnerships.

19. Improvement should not hide inequity in the average

A school average can improve while one group gains less access or another group carries more burden. Disaggregate when there is a legitimate reason and sufficient data to interpret the pattern responsibly. Protect privacy and avoid unstable conclusions from tiny groups.

Ask whether the improvement route was equally usable. Did all intended learners receive the intervention? Were some unable to attend? Did an assessment change introduce an irrelevant barrier? Were students already near the target more likely to benefit because the starting assumptions suited them?

These questions connect school improvement to Educational Equity. A programme is not fully understood until the school knows who could use it and under what conditions.

20. Sustaining means preserving the mechanism, not freezing the first version

EEF’s implementation process includes a sustain phase after explore, prepare and deliver. Sustaining does not mean never changing the programme again. It means maintaining the practices that appear necessary while continuing to monitor context and outcomes. Source: EEF implementation process.

Build the change into ordinary systems: induction for new staff, resource maintenance, timetable allocation, moderation and periodic review. If the approach depends entirely on one champion remembering to push it, it has not yet become institutional capability.

Allow sensible adaptation while checking active ingredients. A school that cannot distinguish adaptation from drift either becomes rigid or gradually loses the mechanism.

21. Stopping can be an improvement decision

An initiative should be allowed to end. Stop or redesign when the educational benefit is not supported, the mechanism proves implausible, the burden is disproportionate, a better option becomes available or the original problem no longer exists.

Stopping does not necessarily mean the people involved failed. A disciplined system learns from negative or ambiguous evidence. It records what was tried, what conditions were present and what should not be repeated automatically.

Celebrate accurate learning about the organisation, not only programmes that survive. Otherwise staff will learn to defend initiatives rather than evaluate them.

22. An illustrative 90-day improvement review

The following is a planning example, not a universal timetable. Days 1–20: define the problem, inspect baseline evidence, identify competing explanations and select a response. Days 21–40: prepare resources, professional learning, roles and measures. Days 41–70: deliver on a bounded scale, inspect implementation and repair obvious friction. Days 71–90: review learner evidence, burden and unintended effects, then decide the next cycle.

For slower-moving outcomes, ninety days may be too short. For an operational routine, it may be longer than necessary. The value of the example is the sequence: diagnosis before purchase, preparation before launch, implementation evidence before overclaiming impact, and a real decision at the end.

At every review, state what remains unknown. A school that can say “the practice is now consistent but we do not yet know whether transfer has improved” is in a stronger learning position than one that converts early enthusiasm into a claim of proven success.

23. A compact improvement brief

Problem: what observable learner or system difficulty are we addressing? Evidence: what tells us this problem exists? Mechanism: how should the proposed action help? Active ingredients: what must be preserved? Infrastructure: what time, training, resources and roles are required?

Implementation evidence: how will we know the practice occurred? Outcome evidence: what would count as educational improvement? Burden: what work is added or displaced? Equity: who may find the route difficult to use? Decision: when and how will we adopt, adapt, expand or stop?

If the brief cannot answer these questions, the project may not be ready for scale. Clarifying them is often cheaper than correcting a poorly understood rollout later.

24. The school itself has to learn

School improvement is education applied to the institution. The organisation has a starting state, forms hypotheses, encounters evidence, makes mistakes, receives feedback and revises what it can do. A school that demands learning from students while refusing to revise its own practices is operating two different standards of evidence.

The goal is not permanent change for its own sake. Stable, effective practice deserves protection. The goal is the ability to distinguish what should remain from what should change—and to make that decision with better evidence than fashion, fear or habit.

Improvement becomes credible when the school can explain the problem, show the change in practice, present evidence of what learners can now do, name the costs and uncertainties, and state what it will change next.

Sources and further reading

Source pages were checked on 5 September 2026. The reading-improvement case, 90-day structure and compact brief are original explanatory designs, not evaluated programmes.

• Institute of Education Sciences: Continuous Improvement in Education—A Toolkit for Schools and Districts, 2020.
• Education Endowment Foundation: A School’s Guide to Implementation, Third Edition, 2024.
• EEF: A structured, but flexible, implementation process.
• EEF: Review of evidence on implementation in education, 2024.