In March 2025 the Royal Statistical Society (RSS) published its statistical analysis of EPSRC’s portfolio. This looked at research grants to better understand differential participation and outcomes by demographic characteristics within the grant funding system.
I really enjoyed being involved with this groundbreaking work as chair of the project steering committee. The Engineering and Physical Sciences Research Council (EPSRC), as far as we’re aware, is the first research council in the world to share its data from each step in the funding process with an independent body for review. This is a really positive step in understanding and eradicating barriers to equality in research.
There is a lot of existing literature about bias in peer review, and EPSRC has previously published reviews of its process with a view to understanding the impact of demographic characteristics on outcome. However, we felt an updated analysis undertaken by statistical and data experts from the RSS would add important insights to the review process.
By bringing in statistical techniques, it was possible to understand the impacts of demographic characteristics at different stages in the review process while controlling for effects at other stages and for other characteristics of grants.
In collaboration with The Alan Turing Institute, it was also possible to undertake a survey of investigator perceptions of bias in the review process. This was by using machine learning methods to analyse demographic trends in reviewer comments.
Dimensions of difference
Sex of investigator was a large focus of previous work by EPSRC, and to a smaller extent ethnicity. We wanted to look into those again but also look at other dimensions of difference that had been less studied. These included disability, institution, region, research area group and theme, and funding mode.
Some of the headline findings are sobering. Ethnic minority researchers had odds of success 32% lower than those of white researchers. Furthermore, ethnic minority researchers when successful in their bids applied for less funding than their white counterparts. For every £1 a white applicant receives, an ethnic minority applicant receives 90p.
In the case of sex, the results were more nuanced. Female fellowship applicants have odds of success 80% higher than male applicants (the difference in odds of success for research grant applications, which make up the bulk of applications, was not significant by sex). However, successful male applicants were more likely to apply for and receive more funding, £1 for every 85p awarded to a female.
Complex findings
Looking into perceptions in the research community also resulted in some complex findings. While ethnic minority researchers were the most likely group to perceive bias, the most common perceived bias was based on the investigator’s institution rather than ethnicity or any other individual demographic characteristic.
In a preliminary analysis of institutional success rates, we did not find clear evidence of bias on the basis of institution, but more research is required before coming to a strong conclusion.
When looking at reviewer scores, there were modest but significant effects of reviewer and applicant ethnicity. In particular, reviewers with white and Chinese ethnicity appeared to give higher scores to applicants that shared their ethnicity.
Other reviewer-applicant ethnicity interactions were not significant, but this could reflect a lack of data for some of the smaller groups.
Principles of responsible statistics
Naturally, given the very sensitive nature of the data involved in this work, EPSRC and the research team made a major effort to ensure that the identities of individuals involved in the review process, whether as applicants, reviewers or panellists, were protected. The dataset made available for research was pseudonymised, with no directly identifying information about any individual included.
Due to concerns that individuals could be identified from combinations of other fields, the names of regions, institutions and research area groups and themes were also masked. In compiling results for publication, care was also taken not to publicly report summary statistics for small groups.
The steering group also followed principles of responsible statistics when determining what analyses should and should not be undertaken, whereby we only proposed analyses if preliminary examination of data indicated that its quantity, quality or nature would permit a robust analysis.
These two considerations are the reason we have not included analyses by ethnicity at greater granularity than are presented in the report, and did not undertake more detailed investigations of regions or subject area groups, some of which had overall outcomes with different characteristics from EPSRC as a whole.
Community engagement opportunity
So, what next? As a first step, members of the community are encouraged to send comments on the report to EPSRC via the Have Your Say: RSS Report survey. EPSRC will also be considering policy implications of the work.
One real takeaway from the work was that the composition of the review panel matters. EPSRC has maintained what they call a ‘mixed gender panel policy’ since 2016, and we found that female applicants were ranked 8.5% higher when at least one panellist was female. More work is needed here on whether this type of policy should also be considered for ethnicity.
The report additionally makes a number of recommendations for future areas of research. Although it explores questions like the connection between prior and subsequent funding success, we can’t determine with this dataset whether this is due to researcher quality, for instance. Future work could help answer that question. Qualitative and experimental methods can complement the quantitative work done here.
To have the best research we need a diverse pool of researchers applying with as much certainty as possible that funding decisions are fair. While there is more work to be done, this groundwork is an essential and important step, and we hope it will be of use to other funding bodies as well.
Dr Eugenie Hunsicker is:
- Chair of the Royal Statistical Society project steering group
- Lead Data Scientist at the Access Group
- Visiting Professor at the School of Mathematical and Computing Sciences, Heriot Watt University