Owning What Can Quit
Douglas J Cumming
Abstract
Academic journals are part of the infrastructure of science, yet most journal titles are owned by commercial publishers while their standing rests on editors and boards who can leave. This paper asks what that separation means for the governance of science. Using a coded dataset of all 57 editorial mass resignations recorded by Retraction Watch from 2015 to 2026, data on editor tenure at 110 economics, finance and accounting journals, and journal-level publication and revenue data, I document three patterns. First, resignation does not reverse an owner’s decision: in none of the 19 episodes involving an editor’s replacement did the owner back down after the board resigned, so editorial communities have influence only before they leave. Second, whether a departing board can take a journal’s standing with it depends on institutions outside the board: successor journals follow most often where a scholarly society backs the board and rarely in medicine, where non-profit hosts are scarce, while journal ranking lists, which rate titles rather than the people who edit them, keep reputation with the owner. Third, a stated limit on editor tenure was applied at the publisher’s discretion; where it was applied, output grew almost fourfold, a short-run gain whose long-run effect on the journal’s standing is not yet known. I draw implications for scholarly societies, funders, publishers, and the bodies that compile journal rankings.
Review reports
xPeerd (AI review)
AI Review (xPeerd — DAReview)
Summary
This manuscript, titled Owning What Can Quit, examines governance conflicts in academic journal publishing through a property-rights framework applied to editorial board resignations and successor-journal formation. It assembles a coded dataset of 57 mass resignations recorded by Retraction Watch between 2015 and 2026, adds economics and finance cases, and combines these with editor-tenure data and journal-level publication/revenue evidence. The manuscript’s main strengths are a clearly stated research question, a coherent theoretical framing, transparent prediction structure (P1–P5), and unusually concrete institutional detail about journal ownership, societies, and ranking systems. The core empirical patterns are also stated plainly in the abstract, including the claims that owner reversals occur before, not after, resignations, that successor formation depends on outside options such as society backing, and that a stated tenure cap did not bind in practice .
The main weaknesses concern evidentiary limits, causal identification, and possible overreach in several interpretations. The manuscript repeatedly acknowledges small samples, imperfect coding, inferred ownership, and author involvement in one focal case, yet some claims remain framed more strongly than the empirical base comfortably supports, especially where journal-level inference relies on very few events or where theoretical constructs are mapped onto coarse proxies . Overall, the paper is thoughtful and potentially publishable, but it would benefit from substantial tightening of claims, clearer separation of description from inference, and more disciplined handling of robustness, selection, and conflict-of-interest concerns.
Potential Major Revisions
- The manuscript should more clearly distinguish descriptive evidence from causal or quasi-causal inference.
The abstract states, “Using a coded dataset of all 57 editorial mass resignations recorded by Retraction Watch from 2015 to 2026, data on editor tenure at 110 economics, finance and accounting journals, and journal-level publication and revenue data, I document three patterns” and then presents broad conclusions about reversals, successor formation, and tenure-cap enforcement . Later, however, the paper itself concedes substantial empirical constraints: “the samples are small,” “the estimates in Tables 3 and 4 should be read as descriptive,” “the journal-level estimates in Table 7 rest on two events,” and “the empirical counterparts are imperfect” . This is an important tension. I think the present draft often oscillates between careful caveating and stronger inferential language than the data warrant.
In particular, statements such as “the evidence shows how complementary institutions… govern the outside option of the human capital” in Section 1/2 framing are stronger than the admitted descriptive character of the design . The revision should systematically relabel the paper as primarily descriptive-comparative unless stronger identification is added. At minimum, each major claim should be tagged as one of: descriptive pattern, model-consistent comparison, or causal estimate. That would materially improve epistemic clarity.
- The mapping from theoretical constructs to empirical proxies is intellectually plausible but often too coarse to bear the weight of the conclusions.
Section 4.3 maps θ, F, G, and C to observable features such as society affiliation, field, volume pressure, fees, editor replacement, and editorial interference . This is elegant as a heuristic, but several mappings compress multiple mechanisms into a single proxy. The manuscript itself admits that “a society can both raise θ and lower F, so the ownership comparisons do not separate the two, and the field comparison may reflect differences in indexing and norms as well as in hosting” . That admission substantially weakens the force of tests presented as evidence on P2 and related propositions.
For example, Section 5.3 states that “Where a society was affiliated with a title owned by the publisher, a high-θ case, a successor followed in six of seven episodes (86 percent)” and contrasts this with “one of eight” society-owned cases and low rates in medicine/life sciences . These are interesting comparisons, but the paper should avoid treating them as clean empirical counterparts of distinct model parameters. A stronger revision would either:
a) explicitly redefine these as bundled institutional environments rather than single-parameter proxies, or
b) perform additional sensitivity analysis showing how conclusions change under alternative codings.
- Selection issues are discussed in theory but not fully absorbed into the empirical rhetoric.
The manuscript is aware of selection problems. Lemma 1 is introduced specifically to justify comparisons “among journals at which the owner proceeds” and acknowledges that the result “fails if the owner knows more about F than H conveys” . Yet the empirical sections still sometimes read as though observed episodes transparently reveal the underlying process. Section 4.1 also notes that the Retraction Watch list “depends on press coverage” and that ownership often had to be inferred from public records . Section 5.1 further states that publisher counts “should not be read as rates” .
These are not minor caveats. They go to the heart of external validity. The revision should make the sample-construction consequences much more visible in the main argument. A stronger version would add a dedicated subsection called “Scope conditions and sample selection,” clarifying that the paper studies publicly visible governance conflicts, not the universe of editorial disputes. As written, the manuscript sometimes implies broader governance regularities than the observable sample can sustain.
- The tenure-cap argument requires more cautious framing because the evidentiary basis appears thin and partly asymmetrical.
Prediction P3 is that “A cap on tenure enforced at the owner’s discretion will not bind” . Table 8 later summarizes this as “binding cap rejected” and “Non-binding: supported” while also noting that selectivity evidence is directionally consistent but not significant . Section 4.4 explains that tenure data come from publicly documented editorial histories, that only 26 Elsevier and 10 comparison editors meet the criteria, and that “Publisher websites were not accessible for verification, so coverage is best for journals whose editorial histories are documented elsewhere” .
This creates a risk that the sample is shaped by documentation visibility rather than a neutral sampling frame. The paper should therefore be much more restrained in treating P3 as strongly supported. The quoted formulation “the only documented non-renewal was at the lowest-revenue journal” also reveals how dependent the inference is on documented cases rather than all cases . I think the revision should either downgrade the claim to suggestive evidence or expand the archival basis sufficiently to justify stronger language.
- The FRL case is potentially informative but currently too exposed to concerns about author involvement, asymmetry of evidence, and overinterpretation.
The paper candidly states, “I was a participant in the FRL transition and have had a dispute with the publisher arising from a different matter” . That disclosure is necessary and commendable. However, because FRL also anchors some of the manuscript’s strongest applied conclusions, including the claim of “almost fourfold” output growth after 2015 in the abstract and the later argument that the owner captured short-run gains , the manuscript faces a heightened burden of methodological distance.
Section 9.1 further strengthens interpretive claims by invoking a citation-network ecology in which “the four most-cited authors in the cluster made 1,006 self-citations and 803 citations to each other’s work” and by suggesting that growth “raised measured impact more than standing” . Even though the paper notes that Alexander and Cumming “do not attribute intent,” the surrounding prose pushes readers toward a critical interpretation. I think the manuscript should separate what is directly shown from what is inferentially suggested. If FRL remains a centerpiece, a revision should either add a more formal protocol for case-study evidence or soften normative implications.
- Some generalizations extend beyond what the empirical base can presently support.
The conclusion-style language in the abstract and late sections sometimes broadens from journal governance to “the governance of science” and even to other organizational settings such as “software projects and newsrooms” . Section 11.3 explicitly says “The setting is narrow, but its structure is common” . The analogy may be intellectually stimulating, but the empirical analysis is highly domain-specific and built around a hand-coded set of visible disputes in scholarly publishing. I think the extrapolation should be presented as conjectural implication rather than demonstrated transportability.
- Anti-fraud / coherence assessment
I do not see strong paper-mill traits such as meaningless jargon strings, fabricated-looking references, or obviously incoherent section progression. The manuscript is conceptually coherent, internally cross-referenced, and candid about limitations and author position. That said, there are integrity-sensitive features that warrant explicit editorial attention: heavy reliance on hand-coded public sources, inferred ownership coding, inaccessible publisher websites for verification, and author participation in one of the central case episodes . These do not indicate fraud, but they do create elevated reproducibility and impartiality risk. The paper would benefit from a replication appendix, codebook, archived source list, and clearer preregistered-style coding rules where possible.
Potential Minor Revisions
- Overstatement in the abstract and highlights should be moderated.
The highlight “A stated ten-year cap on editor tenure did not bind at the publisher citing it” is punchy, but given the later acknowledgment that enforcement evidence partly “rests on a single documented event” and on non-significant patterns by journal grade, the wording should be softened .
- Several formulations would benefit from more precise uncertainty markers.
For example, “Which of the two ends up with the journal’s reputation is not yet known” is appropriately cautious in the introduction, but later passages become more declarative about what ranking lists, societies, or hosting environments do to θ and F . The style would improve if the same caution were maintained consistently across sections.
- Formatting consistency should be checked around symbols and notation.
The manuscript uses θ, F, G, C, R, w, and m consistently in most places, but some readers may benefit from a compact notation table in the main text rather than relying on dispersed explanation across Section 2 and Appendix C .
- Some long sentences could be split for readability.
Example from the abstract: “Using a coded dataset of all 57 editorial mass resignations recorded by Retraction Watch from 2015 to 2026, data on editor tenure at 110 economics, finance and accounting journals, and journal-level publication and revenue data, I document three patterns” . This sentence is grammatically acceptable but information-dense. Dividing the data description from the findings claim would improve readability.
- Minor wording issue: “The people who made the journal worth reading, and the society that lent it standing, are leaving.”
This is rhetorically effective, but the phrase “made the journal worth reading” is somewhat evaluative for an academic article and may read as editorialized rather than analytical . A more neutral wording would better match the paper’s tone.
- Typographic / stylistic consistency in percentages and statistical notation should be standardized.
In Table 3 and related prose, the manuscript mixes textual and symbolic presentations such as “86 percent,” “45 percent,” and “16%” in different contexts, at least in retrieved excerpts . Standardize either worded percentages or symbol percentages throughout.
- Clarify whether “planned,” “announced,” and “launched” successors are always analytically separated.
Section 4.1 says successors are coded as “launched” once announced with a named title and host, and as “planned” when intention is announced without a title , while Table 3 aggregates “launched or planned” . This is reasonable, but the manuscript should explain more clearly whether combining them affects substantive conclusions.
- Footnote density and placement may distract from the main argument.
The footnote discussing exclusion of the Journal of Corporate Finance episode is detailed and useful, but it might be better moved into an appendix or robustness subsection to preserve narrative flow .
- References and source verification
The reference list appears substantively appropriate in the retrieved portions, but because the paper relies heavily on news records, society announcements, and websites, the final version should ensure durable citations, archive links where possible, and access dates for all web sources. The current references already include several access dates, which is good practice .
- AI content analysis
Estimated percentage of AI-generated content: low, approximately 5–15%.
Rationale: the prose is highly topic-specific, argumentatively layered, and includes bespoke institutional detail, field-aware caveats, mathematically integrated exposition, and self-disclosure unlikely to be generated wholesale by generic AI systems. I do not see obvious stereotyped LLM markers such as repetitive generic transitions, shallow paraphrastic inflation, or semantically empty summary blocks. If AI assistance was used, it was likely limited to local editing rather than core intellectual construction.
Potential AI-flagged passages: none strongly suspicious in the retrieved excerpts. The text is too tailored and internally structured to trigger major concern. The epistemic impact of any possible AI-assisted polishing would likely be low relative to the manuscript’s claimed contribution, because the principal contribution lies in dataset construction, model framing, and institutional interpretation rather than generic prose.
Recommendations
-
Recast the central empirical contribution as a descriptive institutional study guided by formal theory, unless stronger identification is added.
-
Add a dedicated limitations subsection in the main text consolidating all caveats now dispersed across sections: public-source dependence, press-coverage selection, inferred ownership, uneven tenure verification, and author participation in FRL-related events .
-
Downgrade or qualify claims of support where evidence is thin, especially for P3 and journal-level inference from two events .
-
Provide a reproducibility package: coding protocol, source archive, excluded-case log, uncertainty flags, and robustness tables using alternative codings for ownership, society affiliation, and successor status.
-
Separate more clearly what the model predicts, what the data describe, and what broader implications are conjectural. This would make the paper more persuasive, not less.
-
Tone down evaluative rhetoric in places where neutral analytical phrasing would better fit the paper’s otherwise serious style.
-
If possible, add additional validation exercises:
- intercoder reliability for hand-coded episodes,
- robustness excluding inferred ownership cases,
- robustness distinguishing launched vs planned successors,
- clearer sensitivity around FRL-centered conclusions.
In its current form, the manuscript is original and intellectually serious, but it still needs substantial revision before its strongest claims can be fully justified by the evidence presented.
Reviewer3 (AI review)
AI Review (reviewer3.com)
Overview
Analysis of Editorial Mass Resignations
Contribution
This study evaluates 57 mass editorial resignations across academic publishing, demonstrating that publishers do not reverse editor replacements after board resignations and that successful successor journals depend on external backing from scholarly societies or non-profit hosts.
Selective Policy Enforcement and Output Growth
Contribution
The author establishes that publisher term limits on editors are applied selectively rather than as binding rules, and that replacing editors to expand publication volume results in rapid multi-fold increases in journal output.
Comments (9)
1. Exit inequality at F = F* + d in Proposition 5 — Editorial
0 in credibility. A pledge changes
In Appendix C.7, the text defines the threshold for exiting after a public pledge using a non-strict inequality that includes equality, but immediately restates it as a strict inequality: "after pledging, the board exits if and only if θwR − F ≥ −C − d, that is, if and only if F < F∗ + d." Since F∗ ≡ θwR + C (from Appendix C.2), the inequality θwR − F ≥ −C − d is algebraically equivalent to F ≤ F∗ + d, not F < F∗ + d. Additionally, in the proof of Proposition 5 on the next page, the text states that "When G ≥ L the owner proceeds anyway, and the board, having pledged, exits with θwR − F ≤ −C", which uses a non-strict inequality for the region F ≥ F∗, whereas the proposition statement itself considers the interval "F∗ ≤ F < F∗ + d".
2. Five 'unclear' reversal episodes treated as zeros without sensitivity check — Major
reversed. The difference (0 of 19 against 3 of 38) is in the predicted direction but too small to be significant (Fisher exact p = 0.54).
In Section 5.2 and Table 2, you test P1 by comparing 0 of 19 reversals in editor-replacement episodes against 3 of 38 in other episodes, but Appendix A codes five episodes (#19 Journal of Biogeography, #26 Publications, #30 Sustainable Organisations section, #35 BMC Pregnancy and Childbirth, and #19 which is an editor-replacement episode) as 'unclear' in the Reversal column. Treating these five 'unclear' outcomes—including one in the editor-replacement group—as zeroes without reporting how they were resolved or conducting a sensitivity check could affect the exact count and comparison.
3. Selection on mass resignations precludes testing pre-resignation voice — Major
Case (b) of Proposition 1, in which voice succeeds, predicts that withdrawals occur before resignations, so they should be largely absent from a list of resignations. The documented case fits.
Because the primary dataset is constructed from the Retraction Watch Mass Resignations List, episodes in which an owner withdraws a contested editor replacement before any mass resignation occurs are systematically excluded by the sampling frame, except when a board happens to resign years later over a different dispute (as at the European Law Journal). Relying on a single incidental case cannot empirically establish how frequently pre-resignation protest succeeds relative to post-resignation protest.
4. Inconsistent coding of unclear ownership and successor status in regressions — Major
Note: Dependent variable equals one if the departing editors or the society launched or planned a successor journal. The omitted ownership category is publisher-owned titles without a society; independent and unclear owners are included in the omitted category.
In Table 4, the note states that the 4 'independent and unclear owners' are pooled into the omitted reference category ('publisher-owned titles without a society') rather than controlled for separately, even though one of those four (#54 Communications on Pure and Applied Mathematics) is society-affiliated (Soc. = 1 in Appendix A) and launched a successor. Furthermore, episode #36 (Gender, Work and Organization) is coded as 'unclear: Feminism and Organization' under Successor in Appendix A, and it is not explained why this is coded as 0 rather than 1 or tested in a sensitivity check.
5. Unmatched comparison between flagship society journal and letters/field journals — Minor
The 2010 ABS guide illustrates both the lag and the way a society can shorten it. It did not rate FRL or the Journal of Financial Stability , both founded in 2004, or the International Review of Economics & Finance , founded in 1992 (Association of Business Schools, 2010). It did rate the JEEA, launched in 2003 with the backing of the European Economic Association, at grade 4 after only seven years, up from grade 3 in the 2009 edition and one grade above the EER, which the association had left. A
To illustrate that society backing shortens the ranking lag on the ABS list, you compare the Journal of the European Economic Association (JEEA)—the flagship general-interest journal of the main European economics society, launched by departing editors of an already top-ranked journal—against specialized or letters finance journals (FRL, JFS, IREF) that were newly founded rather than spinouts from a mass resignation. Because flagship general-interest journals and specialized letters journals differ substantially in scope, citation velocity, and submission quality, this comparison confounds society backing with baseline journal tier and type.
6. Editors reaching ten-year mark late in window had fewer opportunities — Major
vitae, society and university announcements, journal editorials, and RePEc listings (Section 4.4). The test sample consists of editors who had at least ten years of tenure at some point between 2010 and 2015 while in post, the period to which the cap applied; in practice this means spells that began in 2005 or earlier. Both the start and the end of the spell, or the editor’s continuation in post beyond
The binomial test in Table 6 treats all 26 Elsevier editors as having an equal opportunity to be replaced under a 10-year cap during the 2010–2015 window, even though editors who started in 2004 or 2005 (such as Tol, Gençay, Hasan, and Holthausen) only reached their ten-year mark in 2014 or 2015. Because editors serve on multi-year fixed contracts (e.g., three-year terms) and the stated cap applied around 2010–2014, an editor reaching 10 years in 2014 or 2015 might not face a contract renewal decision inside the window or while the policy was active.
7. Asymmetric second-round data collection between Elsevier and comparison journals — Major
The tenure dataset records 549 spells in the top editorial position at 110 journals (65 Elsevier, 45 other), active at any time from 2005 to 2020, collected in two rounds; the second round targeted Elsevier journals with gaps. Each row gives the journal, publisher in 2010, owner type, society
In Section 8 and Appendix B, you note that a second round of data collection targeted Elsevier journals with gaps and nearly doubled the Elsevier sample, whereas the comparison panel of non-Elsevier journals was not extended and remains at only 10 editors across 45 journals. While this does not affect the binomial test within Elsevier, the asymmetric verification effort could introduce differential selection bias when comparing departure rates between Elsevier (8 of 26) and other publishers (1 of 10).
8. Pre-treatment volume surge in 2014 violates parallel trends for 2015 — Major
2013 to $1.86 million in 2018. Figure 4 shows both paths. Two caveats apply. FRL’s output had already doubled in 2014, from 23 to 52 documents, the last year under the outgoing editor. This may reflect a policy that predated the change of editor; in the model what matters is the owner’s policy, so the 2014 rise is consistent with P4, but it makes 2015 a less sharp treatment date, and estimates measured relative to 2014 understate the change relative to 2013; and the window ends in 2018 to exclude the ecosystem, which brought a far larger expansion afterwards (Cumming, 2026a). At the
In Table 7 and Figure 4, you define 2015 as the treatment year for Finance Research Letters, but you note in the text that FRL's output had already more than doubled in 2014 (from 23 to 52 documents) during the final year of the outgoing editor's term. Because the volume expansion began a full year before the new editors took office in January 2015, attributing the post-2014 level shift to the 2014/2015 editor replacement confounds the editorial transition with a pre-existing upward trajectory, and including 2014 in the pre-treatment period masks the true timing of the structural break.
9. Incomplete-information protest is not necessarily cheap talk with pledges — Major
With incomplete information, a pledge is informative because it is costly to a board that would not carry it out: such a board pays d if the owner calls its bluff. The prediction is that successful voice is public and collective, takes the form of commitments to resign, and comes before any resignation.
In Appendix C.7, you assert that under incomplete information a public pledge is informative because a board with a high cost of starting again would not carry it out and would pay credibility cost d if called, yet you do not solve the resulting signaling game or account for how this breaks the pooling assumption in Appendix C.3 and Lemma 1. If boards with F < F* + d pledge at stage 2 while those with F >= F* + d do not, the owner updates its beliefs about F upon observing (or not observing) a pledge, so conditioning on the owner proceeding alters the distribution of F and invalidates the independence argument in Lemma 1.
XSci (AI review)
Summary
Journal titles are alienable assets owned mostly by commercial publishers, while the reputation that gives them value is embodied in editors and boards who can leave. The authors organize the question around a property-rights model of an owner and a board (Section 2, Appendix C) and test five predictions on three bodies of evidence: a coding of all 57 episodes on the Retraction Watch Mass Resignations List, 2015–2026; chief-editor tenure at 110 economics, finance and accounting journals; and journal-level data around two editor replacements. They report that no owner reversed an editor replacement after a board resigned (0 of 19 episodes), that successors follow where a scholarly society backs the board and rarely in medicine, that a stated ten-year cap on editor tenure did not bind, and that output at Finance Research Letters rose about 3.8-fold relative to comparable journals after its 2015 replacement.
Contribution and Significance
The core contribution is the identification of third-party ranking lists as the institution that makes a journal's reputation alienable. Section 3.5 makes the argument precisely: because the Academic Journal Guide, the ABDC list and the FT50 rate titles rather than the people who edit them, a departing board leaves its rating behind, a successor must earn its own over the years a list takes to revise, and authors whose careers depend on the lists have little reason to follow. In the model this lowers θ and raises F at once, and Section 9.1 adds the symmetric consequence: the same lag delays the downgrade of a title whose standing has fallen. Section 11.3 extends the idea convincingly to restaurant guides, legal directories and credit ratings.
What is new is narrower than the framing suggests. The descriptive phenomenon — boards resign, found successors, and struggle to carry indexed status while the vacated title persists — is already documented in work the manuscript does not cite (Major Concern 3). The model is a standard exit-and-voice game whose comparative statics are unsurprising; its best moment is Lemma 1. The distinctive contributions are the owner's-decision framing, the ranking-list mechanism, the tenure-cap test as a transferable design, and the revenue dimension — real, but smaller than the paper claims.
Strengths
- The coding of the 57 episodes (Appendix A) is internally consistent: Tables 2 and 3 reconcile row-by-row with the appendix, and the source-linked dataset will be reusable.
- The field result is robust and new: successors followed in 3 of 19 episodes in medicine and the life sciences against 22 of 38 elsewhere (Fisher exact p = 0.004), surviving the launched-only restriction (p = 0.02).
- Lemma 1 and the discussion after it (Appendix C.3) handle selection squarely, including when the lemma fails and in which direction the bias then runs.
- The conflict of interest is handled unusually well: full disclosure, an explicit revision of the author's own 2014 view (Section 7.1), and a quantified sensitivity analysis for the excluded Journal of Corporate Finance episode (Section 4.3).
Recommendation
Major Revision for Research Policy. The ranking-list mechanism and the coded dataset are contributions the journal should want, and none of the problems below requires abandoning the design. But the tenure-cap finding is contradicted by a pattern in the manuscript's own Table 6, two of the four abstract claims are stated more strongly than the paper's own tests support, and the closest prior work is absent. Each is repairable with data the authors already hold.
Major Concerns
-
The tenure-cap test does not establish what Section 8 concludes (Section 3.2, Section 8, Table 6). Table 6's start and end years contain a pattern the manuscript does not test. Of the eight Elsevier departures, three occurred at exactly ten years of tenure — Gençay (FRL, 2004–2014), Katz (International Review of Law and Economics, 2003–2013) and Hommes (Journal of Economic Dynamics and Control, 2002–2012) — and a fourth, Palivos (Journal of Macroeconomics, 2000–2013), at thirteen. Split at the appointment cohort, Panel A gives thirteen editors starting in 1999 or earlier, who left at 18, 19, 30 and 32 years, and thirteen starting in 2000–2005, who left at 10, 10, 10 and 13. That is what a cap applied prospectively, with incumbents grandfathered, would produce, and it must be ruled out before concluding that the cap "did not bind." The binomial tests do not rule it out: they ask whether editors already at 18, 21, 29 or 42 years in 2010 were removed by 2015, a null no one would hold, and exposure differs sharply (Holthausen reaches ten years only in 2015). A hazard model with tenure as the running variable, and a cohort split, are the right tests and are available in the data already collected.
Section 3.2 also states that the authors are "not aware of a published Elsevier policy limiting the tenure of chief editors," the rule being known from what one incoming editor was told in 2014; the single documented enforcement is the FRL episode, which is also the only informative event behind P4 (the manuscript calls the JBF test "uninformative") and in which the author was an appointed beneficiary. The abstract should say that the rule is attested only by private recollection and that "where it was applied" means one journal.
-
The abstract and Highlights state P1 and P2 more strongly than the paper's own tests (Sections 5.2, 5.3, Table 8). On P2, the abstract says successors "follow most often where a scholarly society backs the board," but Section 5.3's own check shows the society-affiliated share falling to four of seven against 17 of 38 (Fisher p = 0.69), and the coefficient to 0.18 with a standard error of 0.17, once announced-but-unlaunched successors are excluded; the authors concede it "depends on whether announced successors materialize." On P1, "resignation does not reverse an owner's decision ... so editorial communities have influence only before they leave" is two claims. The first is well documented. The second cannot be identified from a list selected on resignation having occurred — as Section 5.2 itself observes, such a list is by construction a list of cases (a) and (c) — and rests on one episode, the European Law Journal in 2018, whose board resigned anyway in 2020; the discriminating test gives Fisher p = 0.54. Table 8 nonetheless grades P1 "Supported."
-
The evidence base for P1 and P2 is one coder's judgment, with AI first passes and no reliability check, and the closest prior work is not engaged (Section 4.1, Appendix A). The codes involve judgment — ownership is "often not stated in public sources" and is inferred, "planned" is distinguished from "launched," three reversal codes are "unclear" — and the disclosure states that first-pass codes for all 57 episodes were AI-produced. No second coder, agreement rate or reliability statistic is reported. Since P2 turns on seven episodes and on two announced successors, independent double-coding of ownership, society affiliation and successor status is needed.
Two papers on precisely this phenomenon are absent. Siler, K. (2026), "Splitting status and substance: The diverging fates of breakaway and zombie journals," Research Policy 55(8), 105519, applies difference-in-differences to breakaway/zombie journal pairs, finds that vacated titles lose citations and author status, and proposes provisional recognition for breakaway journals — the recommendation made in Section 11.2. Van Walsum, S., Matthias, L., Alperin, J. P., and Haustein, S. (2026), "When Editors Revolt: Characterizing Journal Declarations of Independence," arXiv:2606.30394, assembles 39 breakaway pairs since 1989, a longer frame than the Retraction Watch list. The authors should position against both.
Minor Comments
- The Table 6 tests should allow for heterogeneous exposure: Tol, Gençay and Hasan reach ten years only in 2014 and Holthausen in 2015, yet all are scored on whether they left by 2015.
- The FRL placebo p of 0.06 is the floor attainable with 15 controls, so the full placebo distribution should be reported rather than a p-value.
- FRL's documents doubled in 2014 from 23 to 52 and Figure 4 normalizes to 2014; an event study normalized to 2013, reconciled with the synthetic control's pre-2015 RMSPE of 0.16 log points, would help.
- Table 4's robust standard error for a seven-observation cell with six ones is uninformative, as the manuscript says; exact logistic or permutation inference is the fix.
- Proposition 4 predicts bimodal departures, but the share of each board that resigned is not among the codes in Section 4.1, so it is illustrated by GEB and FRL rather than tested on the 57 episodes.
- Brogaard, J., Engelberg, J., and Parsons, C. (2014), "Networks and productivity: Causal evidence from editor rotations," Journal of Financial Economics 111(1), 251–270, is the closest design precedent for Section 9.
- The incoming editors' note of November 2014 identifies the owner's gain G as volume and so identifies P4; since the author co-wrote it, depositing it would let readers verify that identification.
- Proposition 5 and its proof both state the condition as "F ≤ F < F* + d," presumably for F* ≤ F < F* + d, and Appendix C.3 writes "H(F*) ≤ G \ L" for G/L.
- Section 5.1 concludes that editor replacement is "not confined to one publisher" from counts the same paragraph says "should not be read as rates"; normalizing by titles owned would make the claim testable.