SCIENCEYE BLOG

Contact

Directed Acyclic Graph: Find the Research Gaps Worth Funding

8 min read

A glowing network of nodes joined by arrows floats above an open journal while a hand draws one new orange line into it beside drifting fog

SHARE:

By Dr. Lee Stoner

Most researchers draw a directed acyclic graph to pick covariates, then put it away just before it becomes useful.

A directed acyclic graph (DAG) is a causal diagram: each arrow claims one variable may cause another, and each missing arrow claims it does not. In a review of 234 health studies that used DAGs, only 21% stated the causal effect they set out to estimate, and 48% reported the adjustment set their diagram implied.1

Built from the evidence and read with causal algebra (the rules for reading a diagram), the same diagram shows which untested questions your data can answer: your research agenda.

What you will learn What you can do with it
How to build one causal map from your field’s studies Turn a reading list into a diagram you can defend
Why causal algebra matters Tell an answerable gap from one that needs new data
How to rank the gaps Draft a research agenda you can take to a funder

Draw the directed acyclic graph from the evidence

You do not have to invent the arrows. Evidence Synthesis for Constructing Directed Acyclic Graphs (ESC-DAGs) maps each study’s conclusions into its own diagram, checks and corrects every arrow against causal inference principles, and merges the results into an integrated map.2 Dijk and colleagues argue that the diagram should be registered before a systematic review starts, so it can sharpen inclusion criteria and data extraction.3 Stoner and colleagues placed arterial stiffness as the intermediate outcome between the 24-hour activity cycle and cardiovascular disease, and the map below builds on that model.4 Because every arrow traces to a source, each missing arrow becomes a visible question that others can check.

Sketch of a causal map for sitting and heart disease, with stress unmeasuredSketch of a causal map linking age, unmeasured stress, sitting, arterial stiffness and heart disease. Long description follows the figure.Which arrows can this cohort answer?AgeStress(unmeasured)ExposureSittingArterialstiffnessOutcomeHeartdiseaseUntested: does stress change how sittingaffects heart disease? No study found.BoxesBackground cause (age)ExposureAfter exposure (mediator)OutcomeNot measured in this cohortArrowsProposed causeThe questionFrom an unmeasured causeNo arrow: assumed no direct effect.Time runs left to right, and an arrowmeans “may cause”.Sketch, not a validated DAG.
A sketch, not a validated DAG, drawn for a cohort that measured age, sitting, arterial stiffness and heart disease, but not stress. Each arrow comes from the sources in the table below. A DAG cannot show whether stress changes the size of the sitting effect, so that question sits in the note box.
ArrowWhat the evidence showsSource
Age to arterial stiffnessSupported: pulse wave velocity rises with age in reference dataReference Values for Arterial Stiffness Collaboration, 2010
Age to sitting; age to heart diseaseNot reviewed hereNone checked
Stress to sittingLinked in most studies; which comes first is untestedChauntry, 2026
Stress to arterial stiffnessLaboratory studies of acute stress only: stiffness risesZieff, 2025
Stress to heart diseasePooled cohorts: high perceived stress, more coronary heart diseaseRichardson, 2012
Sitting to arterial stiffnessShort-term trials, and cross-sectional links with sitting habitsMoinuddin, 2025; Germano-Soares, 2018
Sitting to heart disease (the question)Pooled cohorts, mostly self-reported sittingBiswas, 2015
Arterial stiffness to heart diseaseStiffness predicts cardiovascular eventsBen-Shlomo, 2014

Evidence behind each arrow, from study abstracts checked on 10 November 2026. Not reviewed here means no source was checked, not that no evidence exists.

A map built from the literature makes the field’s assumptions visible, so you can argue with them.

Why causal algebra matters for your research agenda

Some gaps stay closed however much data you collect. When an unmeasured variable causes both the exposure and the outcome, a larger sample only estimates the wrong effect more precisely. Causal algebra, the rules for reading a diagram, shows which gaps are like this before a grant is spent on them.

The test is identifiability: whether an effect can be computed from variables you could measure, given the arrows you assume. Pearl frames causal inference as answering causal questions from data combined with assumptions, and the diagram is where those assumptions are written down.5

Three rules then run the test. d-separation says whether a path between two variables is open or blocked. The back-door criterion says which variables to adjust for so only the causal path stays open. The do-calculus rewrites the effect of an intervention in observed terms; when no rewrite exists, the gap needs a new measure or a new design.5

RuleWhat it asksWhat it tells you about a gap
d-separationIs there an open path between these two variables?Which patterns your data should show if the map is right
Back-door criterionWhich variables close every non-causal path into the exposure?What you must measure to answer it
Do-calculusCan the effect of an intervention be written in observed terms?Whether your data can answer it, or it needs a new design

Three rules that sort the gaps. Source: Pearl.5

Causal algebra tells you, before you collect anything, whether a gap is answerable.

From missing arrows to a research agenda

Take the map above. In laboratory studies, acute psychological stress raised central pulse wave velocity, a measure of arterial stiffness, by a moderate amount (standardized mean difference 0.51, 95% CI 0.34 to 0.68).6 Across six cohorts, high perceived stress carried a higher risk of coronary heart disease (risk ratio 1.27, 95% CI 1.12 to 1.45).7 Stress therefore opens a back-door path into both sitting and stiffness, so in this cohort neither the sitting to heart disease arrow nor the stiffness to heart disease arrow can be estimated, however large the sample. If stress is measured, adjusting for age and stress closes every back-door path into sitting, while adding sitting closes those into stiffness.

However, the map also has a limit. Whether stress changes how strongly sitting affects heart disease is a question of effect modification, which a directed acyclic graph cannot draw, and a PubMed search on 10 November 2026 found no study that tested it. That gap needs its own design, with stress measured and the sitting effect compared across stress levels.

Sorting arrows into three groups to build an agenda Arrows sort into supported, answerable and needs a measure. The answerable group becomes a ranked research agenda; the needs-a-measure group becomes measures to add. Supported Answerable Needs a measure Your research agenda 1. The gap that unblocks most 2. The next most 3. Measures to add, ranked the same
The middle group becomes the agenda; the third group tells the next grant what to measure.

Sort every arrow on your own map this way: supported, answerable with your data, or waiting on a measure. Sultana and colleagues also merged 35 tobacco cessation studies into one map and found few arrows supporting cultural tailoring, minority stress, and stigma.8 Then rank the gaps by how many questions on your map each one unblocks: here, measuring stress unblocks two.

Ranked, the answerable gaps are a research agenda you can defend.

Takeaway

A directed acyclic graph built from the evidence does two jobs. It shows which untested arrows your current data can answer, and it names the measures you need before the rest can be answered. Evidence and gap maps show where studies are missing.9 A causal map adds which of those gaps your data can close. This week, draw the map for your own question from the studies you already cite. Then mark each arrow as supported, answerable with your data, or needing a new measure, and you have the first draft of a research agenda.

“A practical, systematic and transparent approach for developing DAGs from background knowledge.”

Ferguson and colleagues, describing ESC-DAGs, International Journal of Epidemiology.2

Leave a Reply

Your email address will not be published. Required fields are marked *