How Design-Based Implementation Research Works | Improve an Educational Idea and the System That Must Carry It

eduKateSG Learning Node Series · 0251

An educational programme works beautifully in twelve classrooms.

Then the system tries to use it in twelve hundred.

The materials arrive late. Teachers have less preparation time than the pilot teachers had. Timetables do not leave room for the collaborative work the programme assumes. School leaders interpret the purpose differently. Local communities need adaptations the original designers never anticipated. Professional learning becomes a one-off briefing. Data systems cannot show whether the most important learning processes are happening. After two years, people conclude that the programme “doesn’t scale”.

But what exactly failed? The educational idea—or the system that was supposed to carry it?

Design-Based Implementation Research, or DBIR, begins from the premise that this separation is often false. Educational innovations live inside systems. If an innovation requires new professional knowledge, coordination, leadership routines, data practices, materials, schedules or relationships, those supports are not peripheral implementation details. They are part of the design problem.

DBIR therefore studies and improves the intervention and the infrastructure around it through long-term collaboration among researchers, practitioners and other relevant stakeholders.

Quick answer

Design-Based Implementation Research works by organising long-term research-practice partnerships around persistent educational problems. Partners define the problem together, iteratively design and adapt solutions, study both learning and implementation systematically, and build the capacity needed for improvement to continue after one project or grant ends.

The classic DBIR literature identifies four core commitments: focus on persistent problems of practice from multiple stakeholders’ perspectives; iterative collaborative design; systematic inquiry that develops knowledge about both learning and implementation; and capacity building for sustaining change in systems.

The modern LearnDBIR formulation adds an especially useful emphasis: teams do not only design innovations. They also redesign the infrastructures needed for equitable, effective and sustainable implementation.

The owned reader job

This Learning Node owns the question: how can educational research and practice jointly redesign an innovation and the system conditions needed to make it work, adapt and endure across real settings?

It does not replace How Implementation Fidelity Works, which asks how active ingredients can be preserved while allowing appropriate adaptation. DBIR is broader: it can change the innovation, the supports around it and the partnership’s understanding of the problem itself.

It also differs from How Networked Improvement Communities Work, which owns a networked improvement approach, and from How Knowledge Brokering Works, which focuses on moving knowledge across the research-practice boundary. DBIR is a particular way to organise joint research and development across that boundary over time.

Why “prove it, then scale it” is often too simple

A common model of educational innovation looks linear:

Invent → test → prove effectiveness → distribute → implement at scale.

Sometimes that sequence is useful. But it can hide the fact that the pilot’s success depended on conditions that were never included in the thing being “scaled”. Researchers may have provided unusually intensive coaching. Volunteer teachers may have been especially motivated. Materials may have been adjusted informally by experts. Leaders may have protected planning time. Data may have been reviewed every week.

When the visible curriculum is exported without those invisible supports, implementation weakens. The failure is then blamed on teachers, context or lack of fidelity.

DBIR asks a more demanding question: what whole arrangement made the innovation workable, and how must that arrangement change across settings?

Principle 1: start with a persistent problem of practice, not a favourite solution

DBIR begins with problems that matter to the people living inside the educational system. That sounds obvious, but research projects often begin with a method, technology or intervention looking for somewhere to be applied.

A persistent problem of practice might be:

students can perform procedures but struggle to explain scientific models;
teachers receive assessment data too late to alter instruction;
new curriculum materials work in some schools but collapse where planning time is scarce;
multilingual learners have insufficient authority in classroom knowledge-building;
a professional-learning programme creates short-term change that disappears after facilitators leave.

Different stakeholders can experience the “same” problem differently. Teachers may see workload and lesson time. Leaders may see timetable and staffing constraints. Students may see relevance, identity or access. Researchers may see a learning mechanism. Families may see opportunity and fairness.

DBIR treats those perspectives as inputs to problem definition rather than noise to be removed before research begins.

Principle 2: design collaboratively and iteratively

The first design is not expected to be perfect. Partners build, enact, observe, revise and try again. The iteration is not random tweaking. Each cycle should be connected to evidence and an explicit conjecture about what needs to change.

For example, a new Science unit may assume that teachers can facilitate student argumentation. Early enactments show that the materials contain good tasks, but classroom discussion repeatedly collapses into teacher evaluation of answers. The next iteration may therefore redesign not only the task but also professional-learning routines, teacher guides, observation tools and leadership supports for discussion practice.

The innovation evolves because the implementation evidence changes what the team understands.

Principle 3: study learning and implementation together

Educational research sometimes treats implementation as a nuisance variable: if people had implemented the intervention properly, the “real” effect would be clearer.

DBIR treats implementation as an object of knowledge. Researchers ask not only whether learners improved but how the surrounding system enabled or constrained the work.

Which adaptations improved fit without removing the mechanism?
Which support routines mattered most?
Where did coordination fail?
What knowledge did teachers need to make productive adaptations?
How did leadership, scheduling, materials or assessment interact with classroom practice?
Which local conditions changed the meaning of the intervention?
What became easier after the system learned how to implement?

This produces knowledge about the intervention and knowledge about how educational systems change.

Principle 4: build capacity so improvement can survive the researchers

A project that works only while expert researchers run every meeting has not yet built a sustainable improvement system.

Capacity can include:

teacher knowledge for adapting materials intelligently;
leader routines for protecting time and monitoring implementation;
local facilitators who can support professional learning;
data routines that answer useful questions rather than merely report compliance;
shared design protocols;
institutional memory so learning survives staff turnover;
relationships that let researchers and practitioners investigate new problems together.

LearnDBIR describes capacity for continuous improvement as one of the approach’s four core principles. The point is not to create permanent dependence on external expertise. It is to help the system become more capable of learning and redesigning.

Innovation plus infrastructure

This is one of DBIR’s most useful conceptual moves.

An innovation might be a curriculum, formative assessment process, tutoring model, digital platform or professional-learning design. Its infrastructure includes the arrangements that let people use, learn from and sustain it: schedules, roles, materials distribution, professional learning, assessment alignment, leadership routines, technology, staffing, data systems, policies and social relationships.

If teachers need forty minutes of collaborative planning but the timetable provides none, “teacher fidelity” is not the whole problem. If schools cannot access the required materials reliably, the curriculum is not separable from procurement. If a new discussion model conflicts with high-stakes assessments that reward short individual answers, the assessment system is part of the implementation environment.

DBIR makes those dependencies designable.

Adaptation is not automatically dilution

Scaling discussions often frame adaptation as a threat: the original programme becomes less pure as local actors change it. Sometimes that is true. An adaptation can remove the active ingredient.

But rigid reproduction can also destroy an innovation by ignoring local conditions. The question is not “adapt or maintain fidelity?” It is: what must remain invariant for the intended mechanism to survive, and what should change so the mechanism can function here?

The National Academies’ 2024 report on scaling and sustaining pre-K–12 STEM innovations notes growing evidence for scaling approaches that use principled mutual adaptation and collaborative design, while also calling for better documentation of what is adapted, why and with what effects.

That is a strong discipline for DBIR: adaptation should be visible, reasoned and studied.

A research-practice partnership is not simply researchers being friendly with schools

DBIR often operates inside research-practice partnerships—long-term collaborations organised around educational improvement. The relationship matters because the work requires repeated negotiation of goals, evidence, design authority and local constraints.

Singapore’s National Institute of Education highlighted this in a 2024 SingTeach account of William Penuel’s work on research-practice partnerships. The piece emphasises long-term collaboration, diverse forms of expertise and research that informs ongoing system transformation. It also describes DBIR as one approach used in such partnerships.

Partnership does not mean every participant has identical knowledge or responsibility. Researchers may bring methodological expertise. Teachers bring situated knowledge of learners and classroom constraints. Leaders understand organisational dependencies. Students and communities can identify harms, opportunities and values that formal system data miss.

The design problem improves when those knowledges can challenge one another.

Worked example: scaling a discussion-rich Science curriculum

Imagine a secondary Science curriculum designed around modelling, explanation and evidence-based discussion.

In the pilot, students show stronger reasoning. The ministry or district wants broader adoption. A conventional rollout might print materials, brief teachers and measure outcomes a year later.

A DBIR approach would treat scaling as a continuing design problem.

Jointly diagnose the persistent problem. Is the central problem weak student explanation, insufficient classroom talk, teacher uncertainty about facilitation, assessment misalignment or some combination?
Map infrastructure. What planning time, materials, leadership, assessment and professional learning are required?
Co-design initial supports. Develop curriculum, teacher learning, observation and data routines together.
Study enactment. Observe where classroom learning and system support diverge from the theory.
Iterate. Change materials, supports and infrastructure based on evidence.
Track adaptations across sites. Distinguish productive adaptation from loss of essential mechanism.
Build local capability. Develop facilitators, teacher leaders and organisational routines.
Return findings to theory and design. Explain not only whether the programme worked but under what arrangements and why.

The unit and the system co-evolve.

DBIR is not a licence to change everything until something looks successful

Iteration needs discipline. If every component changes without a record of why, the partnership cannot tell what it learned. Strong DBIR preserves design history.

What problem was the change intended to solve?
What evidence triggered it?
Which mechanism was expected to change?
What was preserved?
What happened after the change?
Did the adaptation help particular groups while harming others?
Should the change become part of the next version?

Iteration should increase knowledge, not merely accumulate versions.

Equity changes the definition of implementation success

An innovation can show a positive average effect while systematically fitting some learners, schools or communities better than others. DBIR’s stakeholder orientation and recent emphasis on equitable implementation make distribution part of the design question.

Useful questions include:

Whose problem definition shaped the project?
Who carries the additional workload?
Which schools have the infrastructure to implement the design as intended?
Who gets to author adaptations?
Which learners gain or lose access?
Does scaling transfer capability to local actors or centralise control elsewhere?

These questions are not decorative values statements. They can change the implementation design itself.

What evidence belongs in DBIR?

There is no single DBIR method. The research design should fit the question. Teams may use experiments, quasi-experiments, observations, interviews, surveys, implementation measures, design records, learning analytics, case studies, network analysis and mixed methods.

The distinguishing feature is organisational: evidence is used inside iterative joint work to improve both knowledge and design. A randomised trial can answer a valuable causal question inside a DBIR programme, but DBIR is not reducible to the trial. Qualitative work can reveal why an implementation breaks, but DBIR is not merely ethnography. Design work matters, but it is tied to systematic inquiry.

A practical DBIR cycle

Negotiate the persistent problem. Include the people who experience and act on it.
Build a shared theory of the problem. Name mechanisms, constraints and system dependencies.
Define the first design hypothesis. What change should improve what mechanism?
Design the innovation and supporting infrastructure together.
Enact in real settings. Preserve variation rather than pretending context disappeared.
Collect learning and implementation evidence.
Interpret evidence jointly. Different stakeholders will notice different failure modes.
Adapt with a traceable rationale.
Build local capability for the next cycle.
Study spread, sustainability and ownership shift. Ask whether the system can keep improving after the original project recedes.

Failure modes

The solution-first failure: the partnership begins with a favourite product rather than a jointly understood problem.
The consultation theatre failure: practitioners are asked for feedback after the important design decisions are already fixed.
The pilot bubble: unusually intensive research support is mistaken for a scalable operating model.
The fidelity-only failure: every local adaptation is treated as deviation rather than evidence about system fit.
The anything-goes adaptation failure: essential mechanisms disappear without being noticed or studied.
The data-without-redesign failure: evidence is collected for publications but does not return to the design.
The grant-dependence failure: the project has no path to local capability, ownership or institutional memory.
The average-success failure: positive mean outcomes hide unequal implementation or access.
The research-practice hierarchy failure: one form of expertise is treated as automatically superior to all others.

For school leaders and teachers

You do not need a formal DBIR project to use its central discipline. When an educational idea fails, avoid asking only, “Did teachers implement it correctly?” Ask whether the system supplied the conditions the idea required, whether those conditions were realistic, which local adaptations were intelligent, and what the implementation taught you about the original design.

When an idea succeeds, do not export the visible lesson materials and forget the invisible infrastructure. Record the planning time, expertise, leadership routines, data, relationships and professional learning that carried the success.

Sources and further reading

Return to the core idea: educational ideas do not travel alone. They travel through people, schedules, institutions, materials, incentives, expertise, data and relationships. Design-Based Implementation Research treats those conditions as part of the thing worth designing and studying. The question is not simply whether an intervention can work. It is whether researchers and practitioners can build a system that learns how to make it work, adapt it responsibly and sustain the capacity to keep improving after the first design is gone.