Why good research doesn’t always reach policymakers

Take two papers addressing much the same policy problem. One appears in a government report; the other leaves no trace in policy.

What separates them?

During his PhD at UCL, supported by Elsevier, Basil Mahfouz developed a way to examine that question at scale. His research found that governments and policy organisations often draw on narrow portions of the available literature. Standard bibliometric indicators were relatively weak at explaining why particular papers were selected. The author and their position within research networks showed a stronger association with policy uptake.

Mahfouz is careful not to reduce policy influence to a single cause. Research use involves institutions, relationships, timing and political context. Still, his findings challenge the assumption that the most useful evidence will naturally rise to the attention of decision-makers.

Comparing the research that travelled with the research that did not

Studying missed evidence presents a basic problem: cited research is visible, while the comparison group is potentially every other paper ever published.

Mahfouz used semantic-matching models to narrow that field. For papers cited by a government or policy institute, the system could identify uncited papers that were closely similar in content. The two could then be compared using bibliometric data, author information and network analysis.

This approach helped separate subject matter from the routes through which research travels. If two papers cover similar ground but only one reaches policy, factors surrounding the work become easier to examine.

The analysis repeatedly pointed towards the role of authors and networks. Researchers with stronger positions in relevant communities were more likely to see their work used. Others produced research that appeared suitable for policy but remained outside the channels through which decision-makers encountered evidence.

Familiarity can become a filter

Sir Geoff Mulgan, Professor at UCL and Mahfouz’s doctoral supervisor, describes a research system producing far more knowledge than its users can readily absorb. “The methods of selection haven’t kept up with that volume,” he says.

Policy teams work under tight deadlines. A familiar expert, institution or evidence source can offer a practical shortcut. National context and language also influence what can be found and assessed quickly.

Mahfouz’s research examined areas including education, climate change and the United Nations Sustainable Development Goals. Mulgan says the work showed patterns of policymakers relying on research from their own country or language, alongside wider biases favouring Global North research over work from the Global South. It also found cases in which potentially relevant evidence from another field had not entered the policy conversation.

None of these filters makes the selected research inherently poor. The concern is the evidence left outside the search.

Research impact also depends on the route

The findings suggest that research institutions need to look more closely at how work travels after publication.

A paper may be accessible online and still remain effectively invisible to a policy team. Discoverability depends on the language used to describe it, the databases and networks through which it circulates, and whether someone can connect its findings to a live policy question.

Mahfouz’s models were also able to identify topics and individual researchers whose work appeared to have policy potential but had received little attention. This could help institutions direct engagement and support towards relevant work beyond their established networks.

New AI tools may widen that search by comparing research according to meaning rather than previous citations or reputation. Mahfouz also notes the limitation: predictive models learn from historical patterns, including the biases within them. Using past policy uptake to anticipate future relevance therefore requires continued scrutiny of what the model may be overlooking.

Better evidence use will still rely on human judgement. A wider and more deliberate search across a broad scientific knowledge base would at least give decision-makers more than the research already closest to their existing networks.