Across 37,073 European Research Council applications, the same proposal scored higher with reviewers who shared the applicant's citizenship and country of work — and that match moved the score about seven times as much as a shared gender did

Reviewers at Europe’s most prestigious research funder give higher scores to applicants who hold their passport and work in their country, according to a study published on 13 August in Nature Human Behaviour. The effect is not a difference between applicants. It is a difference between reviewers reading the same proposal.

Nicola Fuchs-Schündeln and Felix Holub, both at the WZB Berlin Social Science Center, analyzed all 37,073 applications made to the European Research Council between 2019 and 2023. When an applicant and a panel reviewer shared both citizenship and country of work, the reviewer’s first-stage score for the project rose by 0.164 points on the ERC’s five-point scale, which the authors put at 16 percent of a standard deviation. A shared gender moved the same score by 0.024 points.

How the comparison was built

ERC proposals are read by 1 of 27 panels, each organized into groups of around 15 senior researchers whose names stay confidential until after the round closes, the chair excepted. In the first stage, panelists individually score a short version of each proposal assigned to them from 1 to 5, in half-point steps, once for the project and once for the principal investigator. The panel then meets and grades each proposal A, B or C; only an A advances.

Because each proposal is scored by several reviewers and each reviewer scores several proposals, the authors could hold two things constant at once. A fixed effect for the proposal absorbs everything about it, including quality nobody observes. A fixed effect for the reviewer absorbs that reviewer’s general strictness. What is left is the comparison the paper rests on, the same proposal in the hands of a reviewer who matches the applicant versus one who does not.

That design is what lets the authors call the individual-score result causal, and it is the study’s main methodological claim. Earlier work was largely descriptive, comparing success rates rather than scores on identical material.

What sharing a passport was worth

On first-stage project scores, sharing citizenship but not country of work raised the score by 0.075 points (95 percent confidence interval 0.048 to 0.101), and sharing country but not citizenship raised it by 0.056. Sharing both raised it by 0.164. Scores for the principal investigator moved the same way by smaller amounts, 0.083 points when both matched, and second-stage magnitudes were comparable.

The paper’s own comparison is that the single-attribute matches run “on average two to three times larger than for gender.” Set the doubly-matched coefficient against the gender one and the ratio is closer to seven.

The authors then went looking for innocent explanations. If matched reviewers simply share research tastes, controlling for intellectual proximity should shrink the effect. They controlled for semantic similarity between each proposal and the reviewer’s 20 most-cited papers, for direct and indirect coauthorship over the previous decade, and for whether the proposal’s ERC keyword matched the one that came up most often across the proposals that reviewer assessed that year. The coefficients did not move.

If matched reviewers instead know more about the applicant, their scores should be more spread out at both ends. Instead, a matched reviewer raised the chance of a top score without raising the chance of a bottom one. The probability of a perfect 5.0 on a first-stage project score rose by 0.014 when citizenship alone matched, against a baseline of 0.047, while the probability of a 1.0 fell slightly. Better information does not usually work in one direction only.

One thing the authors looked for and did not find: any advantage from shared institutional status. Classifying the ten institutions that win the most grants in each panel as elite, they found no evidence that reviewers favor applicants of their own rank, with every estimate statistically insignificant.

The panel stage is a weaker claim

Individual scores are inputs. Money follows from the panel’s collective decision, and there the evidence changes character: each proposal receives only one panel verdict per stage, the fixed-effect design is unavailable, and the results are correlational. The authors say so.

At the first stage the correlations are there. One additional panel member sharing both citizenship and country was associated with a 1.7 percentage point higher chance of an A, against a baseline of 32.6 percent, and a chair sharing both with 3.1 percentage points, at a p value of 0.055. At the second stage, nothing reached significance for either the A grade or the funding decision.

Combining both stages, the authors estimate that one extra doubly-matched panel member is associated with a 1.0 percentage point higher chance of being funded, against a baseline funding rate of 14.2 percent.

The most concrete version of the gap concerns the five countries that file just over half of all proposals: Germany, the UK, France, Italy and Spain. Their applicants face panels where 4.6 percent of members match both citizenship and country of work, against 1.5 percent for applicants from everywhere else, plus higher shares matching country only (4.5 against 1.8) and citizenship only (3.8 against 2.6). Run those three differences through the model together and the implied funding-probability advantage is 1.7 points versus 0.8, a gap the authors describe as about 6 percent of the baseline success rate.

Gender is the small effect here

The finding likely to be flattened in retelling is the gender one. Same-gender matches did raise individual scores, and the estimates were statistically significant. But they ranged from 2.4 to 4.7 percent of the relevant standard deviations, and the authors state plainly that effects that small are “unlikely to influence final decisions.” They also varied by field, showing up in the life and physical sciences and not reaching significance in the social sciences and humanities.

At panel level the gender result does not merely vanish, it leans the other way. The coefficient on the share of same-gender panel members in second-stage funding decisions was negative, at a p value of 0.081. That is a marginal result on a correlational estimate and should not be read as evidence that same-gender panels punish applicants. It is, however, the opposite of what a headline about gender bias in grant review would lead a reader to expect.

Two economists the ERC hired, on data only the ERC can release

The disclosures matter as much as the coefficients. Fuchs-Schündeln and Holub state in the paper that they worked as contracted experts for the ERC Executive Agency, that the research is “a direct product of this expert engagement,” and that they were paid fees plus travel allowances and reimbursement. Fuchs-Schündeln, who also holds a chair at Goethe University Frankfurt, has been on the receiving end of the process too: she won an ERC Starting Grant in 2010 and a €1.6 million Consolidator Grant announced in November 2018.

The access that made the study possible is also what stops anyone replicating it. The application-level data are confidential under the authors’ expert contract and cannot be published; a researcher who wanted to reproduce the analysis would have to apply to a specific ERCEA unit, and access is at the agency’s discretion. The authors have posted their code with pseudo-data so the pipeline can be run, which is more than nothing and is not the same as replication.

Three further limits sit inside the paper. The panel-level results are associations, so an unobserved pattern — some panels attracting stronger proposals from particular countries while also seating more members from those countries — could produce them. The individual-score effects are averages of a fraction of one scoring increment, and the paper does not show a single grant changing hands. And no correction was applied for testing many hypotheses at once, which the authors disclose.

As of 21 August 2026, eight days after publication, no correction or formal comment had been posted on the article page. The journal names Matthias Egger and Alexander Petersen among its peer reviewers and has published their reports.

What the authors would change

Their proposals are procedural rather than moral. Assign reviewers so that same-country and same-citizenship pairings are limited, as World Trade Organization dispute panels exclude panelists from the countries involved. Diversify panels, which mechanically reduces how often a match occurs.

Or work on the scoring itself. Add a second assessment for proposals near the cut-off, where the authors say small score differences are most consequential. Score the project on materials that strip the investigator’s identity, and score the investigator separately. Or borrow the partial lottery that others have proposed for marginal cases.

The Council, set up in 2007 and running a budget of more than €16 billion from 2021 to 2027, judges proposals on declared scientific excellence, and the paper’s own framing is that homophily works against that goal while handing systematic advantages to applicants from large, well-represented research systems and disadvantages to those from smaller countries. The first stage is where it would bite hardest. It is the gate: a proposal that does not clear it never reaches the interview, the external referees or the discussion where a wrong first read might be corrected.