Peer Review Fails Equity Test Analysis of Submissions to Elife Reveals a Gender Gap in Whom Journals Invite to Do Reviews
Total Page：16
File Type：pdf, Size：1020Kb
IN FOCUS NEWS Micrograph of yttrium studied images of the The technique could eventually help barium copper oxide, pseudogap. Patterns physicists to understand high-temperature a high-temperature in the images, taken superconductivity, says Allan, although he superconductor. with a scanning cautions that the paper is far from definitive tunnelling micro- and that debate about what the pseudogap is scope, often seem disordered to the human eye will continue. because of the material’s naturally chaotic and The work is an impressive, original fluctuating nature, and noise in the measure- application of machine-learning algorithms to ments. The advantage of machine learning in this type of experimental data, says Tremblay. this situation is that algorithms can learn to But the algorithm can only distinguish recognize patterns that are invisible to people. between the various hypotheses it is given, he says, rather than find entirely new patterns. PATTERN RECOGNITION During her talk, Kim said that work is under To train the algorithms, the team fed neural way to apply the technique to rapidly make networks examples of rippled patterns that sense of data from the X-ray diffraction of corresponded to different theoretical predic- quantum materials — a technique that uses the tions. Having learnt to recognize these exam- scattering of electromagnetic waves to reveal ples, each algorithm applied this learning to a mater ial’s 3D physical structure, but which real data from cuprates in the pseudogap. Over creates patterns so rich that they can take 81 iterations, the algorithms repeatedly identi- months to unravel by conventional means. In fied one modulating pattern that corresponded this case, the AI must draw out similarities and to the particle-like description of electrons, classifications itself, rather than be given pre- which dates back to the 1990s. labelled examples, by grouping features that it The team’s paper shows that the particle-like sees as similar. “This journey of using AI, or description is more appropriate in this case machine learning, for various aspects of our they disagree on how to explain these patterns. than is the conventional wave-like description, quest to understand quantum emergence has One approach views electrons as strongly inter- says André-Marie Tremblay, a physicist at the just begun,” said Kim. ■ acting particles2,3, whereas the other treats them University of Sherbrooke in Canada, who was 1. Zhang, Y. et al. Preprint at https://arxiv.org/ as wave-like and only weakly interacting. at Kim’s talk in Beijing. Working out the nature abs/1808.00479 (2018). To glean more information about these of the patterns is crucial to interpreting what 2. Kivelson, S. A., Fradkin, E. & Emery, V. J. Nature 393, patterns, Kim’s team designed neural networks causes them, says Milan Allan, a physicist at 550–553 (1998). — AI inspired by structures in the brain — that Leiden University in the Netherlands. 3. Zaanen, J. Science 286, 251–252 (1999). GENDER BIAS Peer review fails equity test Analysis of submissions to eLife reveals a gender gap in whom journals invite to do reviews. BY DALMEET SINGH CHAWLA journals). In all, the analysis covered the up only 21% of peer reviewers, and around activity of about 7,000 referees, 890 reviewing one in four reviewing editors. Most reviewing omen are inadequately represented editors and 57 senior editors. editors and peer reviewers were based in the as peer reviewers, journal editors The researchers found that women make United States — 62% and 56%, respectively and last authors of studies, accord- Wing to an analysis of manuscript submissions to an influential biomedical journal. PEER-REVIEW PATTERNS The study looked at all submissions made An analysis of thousands of submissions to the journal eLife — in which peer-review panels openly discuss submitted works — found that all-female reviewer groups accepted more manuscripts with female last to the open-access title eLife from its launch authors than did all-male panels. in 2012 to 2017 — nearly 24,000. It found PREPRINT AT BIORXIV PREPRINT AT that women worldwide, and researchers All-male reviewers All-female reviewers Mixed-gender reviewers 80 ET AL. ET outside North America and Europe, were less likely to be peer reviewers, editors and * ns ns last authors. The paper — which hasn’t itself 60 yet been peer-reviewed — was posted on HTTPS://DOI.ORG/10.1101/400515 (2018). the preprint server bioRxiv on 29 August 40 SOURCE: D. MURRAY SOURCE: D. MURRAY (D. Murray et al. Preprint at BioRxiv https:// doi.org/10.1101/400515; 2018). 20 About 7,000 of the submitted studies went 520 1,952 21 56 735 2,211 through the full submission process (at eLife, Accepted submissions (%) 0 authors make a ‘pre-submission query’ before Female Male Female Male Female Male being invited by the journal to send a full paper Gender of last author — a relatively uncommon practice among ©2018 Spri nger Nature Li mited. All ri ghts reserved. ©2018 Spri nger Nature Li mited. All ri ghts reser20ved. SEPTEMBER 2018 | VOL 561 | NATURE | 295 NEWS IN FOCUS — followed by the United Kingdom and in that referees know each other’s identities, Sugimoto says that journal policies should Germany in second and third place. Less than which allows them to discuss any differences aim to ensure diversity on review panels, for 2% of reviewers were in developing nations. of opinion on manuscripts. example, by inviting a greater proportion of Of the full submissions, the study found that women and researchers in developing nations 1,549 (22%) had a female last author — a posi- BODY OF EVIDENCE to do reviews. “This is one of the simplest pol- tion that indicates seniority — and 5,127 had The study is robust, says Jevin West, an infor- icy changes we can make,” she says, “without a male last author. About 53% of manuscripts mation scientist at the University of Washing- high risks, and potentially high benefits.” with male last authors were accepted, com- ton in Seattle. And it is concerning that women Andrew Collings, eLife’s executive editor and pared with around 50% of those with female and authors in developing countries seem to be a study co-author who is based in Cambridge, last authors. marginalized in peer review, he says. “It’s very UK, says that the team is communicating its Fifty-seven per cent of fully submitted important that we have diverse voices repre- results to the editorial board, so that editors can papers with a male last author were accepted sented and that those consider the findings as they assess submissions when the review panel was all male (see ‘Peer- voices are treated “It’s important and select reviewers. “We are particularly keen review patterns’), whereas mixed-gender teams equitably.” that we have to see editors using diverse groups of reviewers accepted 51% of male-last-author papers. And The results echo diverse voices whenever possible.” submissions that had been edited or reviewed previous findings represented and To weed out the effect of implicit biases on by someone in the same country as the corre- about peer review. that those voices acceptance rates, it is tempting to see blinding sponding author were more likely to be accepted This month, a global are treated as a solution, West says. But, he adds, double- than those with a country mismatch. survey by Publons equitably.” blind peer review — in which neither authors The trends are likely to be a result of implicit — a site that allows nor reviewers know each others’ identities — biases, says study co-author Cassidy Sugimoto, academics to record their peer-review activity often works poorly, because some fields are an information scientist at Indiana University — found that researchers in developing coun- so small that reviewers can guess who wrote Bloomington. The study did not seek to reveal tries are under-represented as reviewers, yet a paper. how the disparities arose, say the authors. But are more likely than scientists in richer coun- Sugimoto says that more data are needed because the gender make-up of senior authors tries to accept review requests, and complete to determine the effectiveness of techniques and gatekeepers closely matches disparities reviews faster. such as blinding or open peer review, in which found broadly in science, there is no evidence And last year, an analysis of American reviews are published and authors and review- that eLife is making such disparities worse. Geophysical Union (AGU) journals found ers might know each other’s identities. The research was prompted by eLife, which that women are invited to review less often She hopes that more journals and publishers approached Sugimoto and her colleagues with than expected, but that the editors’ gender has will release data on peer review for analysis. the data; two study authors are eLife employees. no influence on acceptance rates (J. Lerback “Then, we can inform it with evidence rather The journal’s reviewing process is unorthodox and B. Hanson Nature 541, 455–457; 2017). than with anecdote.” ■ ECOLOGY pixels per image as a high-definition (HD) camera, to one of MBARI’s remotely operated vehicles (ROVs). On one of Haddock’s first voyages with the camera, he recorded a Hidden lives of 2.5-centimetre-long animal called an arrow worm emitting a trail of doughnut-shaped rings of blue light. Haddock speculates that the creature uses the display to distract deep-sea animals predators as it escapes. “Our HD camera wouldn’t have captured this at all,” he says. In mid-August, another research team Cameras record behaviours long cloaked in darkness. deployed an 8K camera in the deep sea for the first time to explore hydrothermal vents BY AMY MAXMEN using extremely high-resolution cameras and in the Okinawa Trough near Japan.