David Banks, Duke University
I worry that our journals and publication processes no longer serve the needs of the statistical profession. It seems past time to reconsider things.
At the very least, can we agree to stop publishing hard copies? It is very expensive, it harms the environment, and I don’t think very many people still read journals that way—I know of none. Personally, I find some value in looking at a table of contents; it is good to know what people are thinking about and what my friends are publishing. But I could get that online or from an email notification.
Years ago, when I first advocated that we go entirely electronic, a member of the ASA Board of Directors objected on the grounds that if there were an electromagnetic pulse event, much statistical knowledge would be lost. My view is that if there were an EMP disaster that took out all electronics, we would probably have greater problems than the preservation of theorems about local asymptotic minimaxity. But I take the point: Let’s have the editor of each journal print out a copy of each published paper and store it in a climate-controlled vault.
More seriously, back when I was one of the editors of the Journal of the American Statistical Association, I became concerned about the self-imposed limits we have placed on ourselves. Publication works very much the way it did 80 years ago, but the technology of communication has leapt forward. Here are some of the main problems I see:
- Refereeing is labor-intensive, highly variable, and may provide false reassurance about quality.
- Print media has no effective mechanism for correction—I was embarrassed to publish a corrigendum to a paper JASA published in 1968 because there was no way correction would be discovered by a reasonable researcher.
- Print publications impose artificial limits on length. A good paper may be quite long with many useful examples and detailed proofs, but journals will not publish those.
I have additional concerns, but these three seem sufficient to warrant a discussion about how new publication practices could advantage our field.
I have ideas about how we could change things to improve, but this note is not the place to put those forward. Instead, I urge the American Statistical Association to convene a panel to review our practices and suggest ways to modernize publication. If we decide to change nothing, at least let it be a deliberate decision, rather than passive deference to the past.
Thanks to David for putting this issue so accurately: “I worry that our journals and publication processes no longer serve the needs of the statistical profession. It seems past time to reconsider things.” How we publish is not just a technical detail of operation; it is an essential part of our disciplinary culture and identity — and a powerful lever for change. Current practices in submission, reviewing, and dissemination sometimes affect the ways our discipline contributes to pressing scientific research and societal problems. I see David’s invitation not as one on unpacking every issue but about recognizing opportunities and taking action. By modernizing our practices, we can strengthen the role of statistics in interdisciplinary science, foster transparency and reproducibility, and amplify our contributions to the many areas where rigorous use of data and statistical innovation are urgently needed.
Interesting ideas. I don’t use paper copies anymore and have long wondered why journals still have page limits!
I asked ChatGPT about journal links and it said, “The International Committee of Medical Journal Editors (ICMJE) and COPE (Committee on Publication Ethics) recommend that corrections and commentaries be clearly linked to the original article. Compliance is not universal, but most reputable journals follow it now.” So with electronic journals we can indeed greatly improve how we handle corrections.
One other note — David suggests that journal editors print paper copies. That is a pretty safe method, but makes it hard to quickly recover. I also asked my chatty friend and it said that data on CDs, flash drives and even external hard drives, all of which are in drawers and not connected to anything when an event happens, are pretty safe from solar flares and other electrical disruptions. With a bit of tweaking to the storage they can be made even safer. So maybe each journal should also by a 10 TB external drive and do a complete period backup. Then stick it in a vault.
I wholeheartedly support a move to electronic journals.
Right on David! Your comments about paper copies are impossible to argue with, but I wonder what (likely more controversial) changes you’d suggest to the traditional referee process?
I feel this is a high time for the community to take a closer look at journals and their support systems. I sincerely thank the author for drawing our attention to this important topic.
I would like to highlight two areas of concern:
a) Over the past thirty years, several new journals have been established, all committed to rigorous peer review. However, it has become increasingly difficult to secure timely and thorough reviews. We lack formal processes to train young reviewers, and unlike machine learning conferences, we have no mechanism to require submitting authors to contribute reviews. As a result, we are seeing a wide divergence in the quality and viewpoints of referee reports for the same paper.
b) Applied statistics journals face a particularly serious problem. Review times are often excessively long, and even after repeated reminders, the reports we receive are sometimes not rigorous enough to be usable. If applied journals are to remain competitive with other empirical science journals, they urgently need to improve both the speed of their review processes.
The current peer-review system in the statistics community, which is known for its lack of transparency and openness, has led to concerns about the quality of published research. Some papers contain significant errors, a problem that mirrors issues in machine learning research identified by Schaeffer et al. (2025). They argue that the current peer-review process is inadequate and fails to prevent the publication of “misleading, incorrect, flawed, or perhaps even fraudulent studies,” advocating instead for a “dynamic self-correcting research ecosystem.”
The review process itself could be a major part of the problem. It’s not only labor-intensive and highly inconsistent, but the existence of a few highly flawed papers raises questions about the entire system. Two papers from the Journal of the American Statistical Association, https://doi.org/10.1080/01621459.2024.2412364 and https://doi.org/10.1080/01621459.2021.1999818, contain striking and critical errors in their proofs. Details can be found in a comment on arXiv: https://arxiv.org/pdf/2509.03702.
For example, in the case of Wang et al. (2025), every major result, including Theorem 1, Theorem 2, and Proposition 1, is flawed. A reviewer with even minimal diligence could have caught these errors without an exhaustive review. Similarly, Liu et al. (2023) contain a wildly incorrect lemma as the very first theoretical result in their supplementary materials.
The frequency and severity of these errors raise a fundamental question about the importance and verification of mathematical proofs within the field. Rather than attempting the likely impractical task of reforming the peer-review system, a more viable solution, echoing Schaeffer et al. (2025), would be to create a high-profile, reputable platform—perhaps a “Refutations and Critiques” track on arXiv or within each journal. This would give crucial visibility to research that critically challenges prior work, thereby enhancing the reliability and credibility of statistical research.
This is a great article that addresses a critical issue in the statistics community. I couldn’t agree more that the current peer-review system often falls short, leading to the publication of papers with significant errors. The concerns raised by Schaeffer et al. (2025) [https://arxiv.org/abs/2506.19882] about the challenges in machine learning research echo this problem perfectly. They make a powerful case for a “dynamic self-correcting research ecosystem,” and I think their argument that reforming the peer-review system directly is impractical is spot on.
Instead of trying to fix a system from the inside, a more effective approach would be to build a new, visible platform for critical analysis. I love the idea of a “Refutations and Critiques” track on arXiv or for each journal. Giving a high-profile home to research that challenges prior work would be a crucial step in enhancing the reliability and credibility of statistical research. This kind of open, transparent discourse is exactly what’s needed to move our field forward.