A new OSIRIS–TIER2 study published in Royal Society Open Science examines how interventions designed to improve reproducibility seek to influence research behaviour. Using the Behaviour Change Wheel framework, the review looks at whether these interventions work by requiring, encouraging or enabling changes in research practice.

Building on the OSIRIS–TIER2 scoping review

The paper is a direct continuation of work published by the same project collaboration last year. In their 2025 scoping review, OSIRIS and TIER2 colleagues screened more than 36,000 records to establish which interventions to improve reproducibility have ever been empirically evaluated. They identified 105 studies and mapped them onto evidence gap maps by intervention type, outcome and study design.

The 2026 study asked a different question of that same evidence base. The authors examined the intervention mechanisms and the policy approaches through which they are delivered.  This makes it possible to bring otherwise very different interventions—from journal data-sharing requirements and reporting guidelines to software tools, workshops and open science badges—into a common analytical framework, and to see which approaches dominate the evidence and which remain comparatively unexplored.

Looking at interventions as behaviour change

For this analysis, the researchers used the Behaviour Change Wheel (BCW), a framework originally developed for public health interventions and used widely in intervention research since. BCW groups interventions by their functions—broad ways of influencing behaviour, such as coercion, persuasion, incentivisation, training, modelling or enablement—and policy categories through which interventions can be implemented, including guidelines, regulation, service provision, communication and legislation.

The team classified the interventions according to these two BCW layers. This produced a common map showing which behaviour-change approaches have been evaluated, which policy routes have supported them, and how they are distributed across different reproducibility outcomes. The researchers also considered the direction of effects reported by the original studies.

Those findings were then interpreted using COM-B, the behavioural model at the centre of the BCW, to see which drivers of behaviour the evidence base addresses. COM-B describes behaviour as shaped by three interacting conditions: capability (knowledge and skills); opportunity (social and material conditions); and motivation (processes that energise and direct behaviour).

What the mapping shows

The analysis reveals a clear imbalance. Coercion and persuasion were the most commonly evaluated intervention types, most often implemented through regulations and guidelines. Across the reviewed evidence, mandatory approaches generally showed better results than voluntary ones, particularly for practices such as data sharing. Yet mandates were not automatically effective: policies without meaningful enforcement or monitoring often produced incomplete compliance.

Enablement—providing tools, infrastructure or practical support—was also relatively common and generally showed promising results. By contrast, incentivisation, training and modelling were rarely evaluated, leaving little empirical evidence about how well they work. This comes as a surprise, given their prominence in discussions about reproducibility reforms.

The mapping also shows which stakeholders lead interventions. Journals were the main implementers, while funder- or university-led interventions were rare. This concentration also raises a question of timing: journal interventions usually act at submission, after the design and analysis decisions that determine reproducibility have already been made. Earlier interventions that shape research workflows deserve more attention.

Looking across the findings through the COM-B model, a broader issue emerges. Much of the evaluated evidence concentrates on interventions that influence motivation, particularly through requirements and encouragement. But reproducible research also depends on researchers having the capability to carry out good research practices and the opportunity provided by appropriate tools, infrastructure, time, resources and supportive research environments.

What does this mean for reproducibility policy?

Taken together, the findings show an evidence base concentrated around a relatively narrow set of approaches. Requirements have an important role, particularly where practices would otherwise remain optional, and the evidence suggests that they are more effective when backed by clear enforcement or monitoring. But requirements and encouragement alone are unlikely to produce sustained change. We need interventions that build skills, provide practical support and infrastructure, and strengthen incentives and supportive norms.

This is where important gaps in the evidence remain. Training, incentivisation and modelling are prominent in discussions, yet they have been evaluated far less often.  Future work therefore needs both to evaluate these under-studied approaches and to test more strategically designed interventions that address different components of behaviour together, rather than relying predominantly on a single mechanism.

The BCW mapping applied in the study provides a structure for building that evidence cumulatively. As new interventions are evaluated, they can be classified within the same framework, making it easier to see which approaches have been tested, how they have been implemented, where the evidence remains thin, and under what conditions different strategies appear to work.

The article is available at: https://doi.org/10.1098/rsos.252472
Paper title: Open science interventions to improve reproducibility in science: a narrative review from a behaviour change wheel perspective
Project: OSIRIS – Open Science to Increase Reproducibility in Science & TIER2 – Enhancing Trust, Integrity and Efficiency in Research through next-level Reproducibility
Corresponding author: Magdalena Kozula (magdalena.kozula@kuleuven.be)