Pathform AI Browse packs

Research Roadmaps

What Is a Research Roadmap? The Five Layers Behind a Clear Diagram

Most roadmaps are drawn as flowcharts, and flowcharts invite boxes that look tidy but say nothing about dependency. Thinking in layers instead forces a harder question about each element: which layer does this belong to, and what does it receive from the layer above?

At a glance

Cluster
Research Roadmaps
Reading time
6 min
Format
Concept walk-through
Sources
5 listed

Guide clusterResearch Roadmaps

Direct answer

A research roadmap is a diagram of the reasoning that connects a research question to a defensible conclusion. It has five layers: the question and the gap it fills, the data or materials, the method, the validation criteria, and the translation of results. The layers run in that order because each depends on evidence from the layer before it.

01Layer one: the question and the gap it fills

The top layer is not the research question alone. It is the pair of the question and the specific gap that makes the question worth answering. A question without a gap reads as curiosity; a gap without a question reads as a complaint about the literature. Reviewers look for the pairing, because it tells them what the project will change in the field rather than simply what it will measure.

A workable gap statement is narrow enough to be falsifiable. Claiming that little is known about a broad topic invites the response that plenty is known. Claiming that no study has compared two specific conditions in a specific population over a specific interval gives the rest of the roadmap something precise to reference.

Question layer: which gap does this study close, and why does it matter now?
Inputs layer: what does the project consume, and is each item confirmed accessible?
Method layer: what is converted into what, and by which procedure?
Validation layer: what result would show the main claim is wrong?
Translation layer: who uses the result, and in what form does it reach them?

02Layer two: inputs, and why they come before methods

The second layer names what the project consumes: samples, datasets, instruments, participants, archives, simulations or field sites. Putting inputs above methods looks strange in a flowchart, where the method usually appears first, but it reflects how a study actually constrains itself. Method choices are only meaningful against the material available, and a method box that outruns its inputs is the most common source of unachievable aims.

This layer is also where feasibility is decided. If an archive turns out to be inaccessible, or a cohort too small after exclusions, the method must be rewritten rather than hoped for. Reviewers with domain expertise read this layer first.

If an input is not confirmed accessible, mark it as provisional in the diagram rather than presenting it as settled.

03Layer three: the method as a converter

The method layer explains how one thing becomes another: raw measurements become estimates, a cohort becomes an effect size, a corpus becomes coded categories, a design becomes a tested prototype. Describing the method as a converter rather than as a task list keeps the diagram honest, because it forces you to state both what goes in and what comes out.

When a single study needs several methods in sequence, keep them in separate boxes with labeled connections rather than merging them into one large box. A merged method box hides the dependency between steps and makes it impossible for a reviewer to see where the project could break.

  1. Name the input in the terms your data are actually stored in.
  2. State the operation in the vocabulary your field uses for it.
  3. Name the output together with its uncertainty, not only the procedure.
  4. Confirm that the output of this box is the input of the next one.

04Layer four: validation as a first-class layer

Validation is the layer most often left out or folded into the method. It deserves its own position because it answers a different question. The method layer says what you will do; the validation layer says what would convince a skeptical reader that the result is not an artifact of your choices. That means baselines, controls, held-out samples, sensitivity analyses or independent replication, depending on the field.

A useful test is to write the validation box as a conditional sentence: if the result is real, then the comparison against this baseline will differ in this direction. If you cannot complete that sentence, the layer is empty.

Named baseline or control condition, not a vague reference to prior work.
A stated uncertainty measure, such as a confidence interval or an error bound.
A pre-specified criterion for success, decided before the data are seen.
A documented handling of cases that could bias the result, including exclusions and missing data.

05Layer five: translation, and how the layers stay separate

The bottom layer connects the study to something outside it: a decision, a design, a guideline, a policy input, or the next question the lab will ask. Translation is not the same as an impact paragraph. It should name the user and the form the result takes when they use it, which is usually more modest and more credible than a claim about benefits.

To keep layers from collapsing into one another, run a simple audit on your draft. Each box should be answerable by one question only. What problem does this solve. What does it consume. What does it produce. What would falsify it. Who acts on it. When two boxes answer the same question, one of them is redundant, and when one box answers two, it needs to be split.

Worked example

Worked example: a reading intervention study, layer by layer

An education researcher wants to know whether a structured phonics sequence improves decoding scores more than a balanced-literacy approach for second-language learners in grades two and three. The first layer pairs that question with a gap: existing comparisons bundle several instructional components together, so it is unclear whether the sequence or the additional time on task drives the difference.

Layer two names the inputs as a specific set of partner schools, an agreed pupil sample after consent, a standardized decoding measure with published reliability, and teacher time for training. Layer three describes the method as a conversion: assignment to condition becomes a comparable baseline, the ten-week sequence becomes an exposure measure, and the post-test becomes an outcome estimate adjusted for the pre-test.

Layer four supplies validation through a delayed post-test, a blinded scorer for the oral component, and a planned check that attrition in the two arms is similar. Layer five translates the finding into a placement decision that schools can apply at the start of the year, and into a follow-up question about whether gains persist after the sequence ends. Read in order, the five layers tell the whole study in about ninety seconds, which is the practical target for a review figure.

Common mistakes

Avoid thisDo this instead

Treating layers as a filing system

Use the layers to test dependency, not to sort content; every box should receive something from the layer above it.

A gap statement that is too broad

Narrow the gap to a specific comparison, population and interval so the rest of the roadmap can reference it precisely.

Validation folded into the method

Give validation its own position in the diagram and write it as a falsifiable conditional sentence.

Translation written as an impact claim

Name the user and the form the result takes, and keep the claim to what the study design can actually support.

Frequently asked questions

Do all five layers have to appear in the diagram?

For a full proposal or thesis, yes. For a short internal document, you can compress inputs and method into one box, but keep validation visible, because that is the layer reviewers use to judge whether the design is credible.

How is a roadmap layer different from a project phase?

A phase is a period of time and is usually defined by workload. A layer is defined by the type of claim it supports. Two phases can sit inside one layer, and one phase can span three layers, which is why roadmaps and timelines often disagree.

Is it safe to put unpublished research design into an AI tool?

Check your institution's data policy first. Many tools process uploads on external servers, which may conflict with confidentiality agreements or ethics approval. Remove identifying participant information, and prefer generating the layout from a de-identified outline.

Can the layers be ordered differently for qualitative research?

The labels change more than the order does. In an interpretive study, the method layer describes sampling and analytic strategy, and validation covers reflexivity, triangulation and audit trails. The dependency still runs from question to inputs to method to validation to translation.

Sources and further reading

  1. Academic writing and structure guidancePurdue Online Writing Lab, Purdue University
  2. CRediT - Contributor Roles TaxonomyNISO
  3. LaTeX templates for theses, papers and proposalsOverleaf
  4. EQUATOR Network - library of reporting guidelinesUK EQUATOR Centre, University of Oxford
  5. ORCID - persistent identifiers for researchersORCID

These are the organisations whose published guidance the recommendations on this page follow. Listing them does not imply that they endorse Pathform AI, and this page is not affiliated with them.

Put this guide into practice

Get the editable layouts this guide describes, ready to retype and submit.

Browse template packs