Economics Peer-Review: Problems,Recent Developments,and Reform Proposals

Abstract

This article contributes to the debate on reforming the peer-review process in the economics profession. It examines the current state of peer-review in economics, surveys the relevant literature, and identifies several problems and solutions. Problems to be discussed are referee overreach and excessive revisions, strategic refereeing and conflicts of interest, prestige bias and other discrimination, and the noisy outcome of peer-review. It recommends several solutions for reform. First, enforce referee guidelines that referee reports must explicitly separate their suggestions into essential and optional, with 3 essential maximum. Second, let authors award the best referee report. Third, adopt conflict of interest policies for referees and punish non-disclosure. Fourth, use double-blind refereeing. Fifth, make better use of prior reports from other journals. Sixth, pay referees for prompt reports. A discussion of the role of editors highlights additional issues that deserve a debate in the profession.

Introduction

Many economists agree that peer-review is not working well (e.g., Akerlof, 2020; Frey, 2003; McAfee, 2014; Rubinstein, 2017). Most referees and editors do a great job and are underpaid for what they do; nevertheless, the process often leads to unsatisfactory outcomes. Yet, aside from trying to shorten review times, relatively little has been done about it (Rubinstein, 2017). Azar (2006) argued very eloquently why progress on such issues of the peer-review process is hard to achieve:

In academia, there are incentives for individuals to increase their efficiency, but […] no one has both the power and the incentives to improve the system. While society may benefit from increased productivity in academia, there are no managers with the authority and incentives to make the production process of academic knowledge optimal. It is therefore important that researchers will dedicate time to think seriously about the issues involved in this process.

Indeed, even a well-known economist and former AER and Economic Inquiry editor, who was very aware of many of these issues (McAfee, 2014) and tried a radical remedy at Economic Inquiry, has left university employment in favor of a corporate gig. A single journal or editor cannot fix these issues, as norms and expectations among referees appear to play a role. Thus, without a coordinated move for change in the profession, these peer-review issues will persist or even worsen, as did the time-to-acceptance over the previous decades (Ellison, 2002b).

This paper attempts to make a contribution by formulating the issues, organizing the literature around them, and by discussing possible solutions. The hope is that it helps to bring about positive change in the economics publication process; hence, I focus on providing a concise survey and discussion rather than an exhausting discussion of all possible issues. To that end, I synthesize a short list of realistic recommendations for reform, based on the most promising solutions.

Among the problems to be discussed in this article are referees increasingly asking for more and overreaching changes, thus shifting the original focus of the manuscript, making manuscripts longer, reducing readability and increasing publication times. Further problems include how the peer-review system is biased against certain authors, that conflicts of interest affect referee recommendations and ultimately publication decisions, and that the peer-review outcome is noisy and unpredictable.

To address these issues, I first recommend that journals enforce standards that referee reports must explicitly separate their suggestions into essential and optional, and that the number of essential changes asked for is capped at 3. This limits the scope of revisions, saves everyone (authors, editors, referees) time, and likely leads to less bloated and more readable papers by focusing on the main issues when revising. Second, I recommend that authors can select one report per first submission for an award. The hope is that these awards make refereeing more attractive and reports more useful. Third, I recommend that conflict of interest policies for referees are adopted by all journals, that these include “competing papers,” that failure to disclose conflicts is punished, and that editors refrain from using referees with conflicts. Fourth, I recommend double-blind peer-review, which is easy to circumvent for referees, but evidence suggests it might still be partially effective in reducing biases in peer-review and the implementation costs are small. Fifth, I recommend that journals encourage the sharing of prior reports from submissions at other journals by authors. If a manuscript was narrowly rejected based on a lack of novelty at a top journal, then it would be a waste to redo the review at a lower ranked journal, and editors should be open to using prior reports for a decision.

The good news is that some of these measures are already in use at selected journals. Sharing of prior referee reports is possible within some journal families (e.g., the American Economic Association, AEA), though typically not beyond. And Econometrica is now using ambitious new referee guidelines aimed at avoiding excessive revisions. So attempts at improvements are not unheard of, but still rare.

While slow review and publication times are an issue (e.g., Ellison, 2002b), I will not discuss it at length here, because out of all the problems with peer-review, this one received quite a bit of attention from journals already. ¹ Since other problems discussed here, such as referee overreach or low referee agreement, are partial causes of the slow publication process, solving these issues will also speed up the process. In addition, sixth, I recommend that referees are paid for prompt (within 4 weeks) reviews at all journals that charge authors submission fees. Payment as well as short deadlines have been shown to reduce review times in a convincing field experiment (Chetty et al., 2014), with some preliminary evidence that review quality does not suffer. ²

I will also not discuss the problems concerning p-hacking, publication bias, or non-reproducibility of empirical research (e.g., Brodeur et al., 2016; Brodeur et al., 2020; Camerer et al., 2016). These are very deep issues beyond the referee process. For example, the code and data at most journals do not have to be shared by authors with the submission, so there is not much referees can do to check the data themselves.

There have been previous attempts to discuss and research issues with peer-review, which are surveyed below. A special status belongs to the Evaluating Research project (https://evalresearch.weebly.com/), which ran a survey among about 1500 publishing and refereeing academic economists. They summarized the literature on peer-review, and discussed all proposals for reform in the literature and from their survey in Charness et al. (2022), sometimes with survey evidence on these proposals’ popularity among economists. Theirs is an excellent report, but given over 150 discussed proposals, it might be too extensive for reform action. In contrast, after discussion of promising candidates, I propose a few very straightforward changes, which are hopefully easier to use in order to change policies.

The following sections aim to clearly articulate the current issues of economics peer-review and discuss potential solutions. Often there is no comprehensive scientific evidence on these solutions, in which case I at least mention at which journals these measures have been attempted. The recommendation section at the end selects the most promising solutions as recommendations.

Problem 1: Referee Overreach and Excessive Revisions

Problem Description

What is the task of the referee? Is it:

1. To see if there are errors or omissions in the manuscript, and judge the contribution to the literature (evaluation)? Or:

2. To help improve the manuscript to the best state it can get to within 1–3 revisions (improvement)?

Peer-review in economics started with evaluation in mind, but has over time added on the improvement task (Ellison, 2002b). On the one hand, why not have referees suggest improvements while they evaluate the manuscript, so both objectives are compatible. On the other hand, this task extension today leads to many more suggestions for changes and more revisions. Too often, these suggested changes are subjective, marginal, or actually make the manuscript worse, so they add a large time cost for referees, authors, and editors without actually improving the manuscript. ³ An additional problem is that inter-referee agreement is typically low (see the noisy outcomes section below), so the suggested changes can be idiosyncratic or even contradictory. A former editor of the AER succinctly summarizes some of these problems (McAfee, 2014 with my emphases):

The way that economists operate journals is perhaps the most inefficient operation that I encounter on a regular basis. It is a fabulous irony that a profession obsessed with efficiency operates its core business in an inefficient manner. […] Many hours are devoted to reviewing papers. This would be socially efficient if the paper was improved in a way that is commensurate with the time spent, but revising papers using blind referees often makes papers worse. Referees offer specific advice that push papers away from the author’s intent. It is one thing for a referee to say, “I do not find this paper compelling because of X,” and another thing entirely to say that the referee would rather see a different paper on the same general topic and try to get the author to write it. The latter is all too common. Gradually, […] we have transformed the business of refereeing from the evaluation of contributions with a little grammatical help into an elaborate system of glacier-paced anonymous coauthorship. […] My sense is that the first revision of papers generally improves them and that it is downhill from there.

Frey (2003) has similar concerns, arguing every referee effectively has veto power, thus forcing authors to follow each of their “suggestions.” This also makes it harder for unconventional ideas or approaches to succeed. Akerlof (2020) echoes these concerns: “As I understand it, a ‘review’ is a journal that takes submissions, and decides which to accept/which to reject. That means that the editors and the referees should be viewing themselves as helpmates, rather than dictators holding authors at ransom before accepting their submissions.” If these problems are so pervasive, their consequences should be measurable and visible in the data, and indeed they are.

Ellison (2002b) finds that submission-to-acceptance times at major economics journals have increased from 6-12 months in 1970 to 18–30 months in 1990. The mean time-to-acceptance was 6.1 months in 1970 and increased to 17.3 months by 1999 (Ellison, 2002b, Table 1). A large part of this slowdown in economics publishing is attributed to the increase in rounds of revisions. Before 1970, “revise and resubmit” decisions where rare and exceptional, and manuscripts were either accepted or rejected, and sometimes got an “accept and revise” decision. Today, straight acceptances are almost unheard of, and first round decisions are either rejections or R&Rs. The mean number of revision rounds increased from about 0.6 in 1960 to about 2.0 in 1990 (Ellison, 2002b, p. 956). These delays are even greater now. ⁴

Manuscripts get longer, with more extensions, robustness checks, and qualifications. Between 1970 and 2010, the mean paper length in top economics journals increased from about 16 to 46 pages (Card & DellaVigna, 2013). Part of this increase is anticipation of the peer-review process, part of it is its product, with referees asking for additions. As Rubinstein (2017) notes, it can make papers less readable.

One cause of the problem might be that we are stuck in a “bad equilibrium” (Hadavand et al., 2024; Rubinstein, 2017) and expectations in this equilibrium call for referees to provide at least 4 pages of comments and suggestions, because that’s how their reports look. So they fill those pages, even though they may be happy with the manuscript except for perhaps two things. But to please the editor (in their mind), they turn those two issues into ten. Multiply this by 2–3 referees, plus maybe an associate editor, and a revision has to address 20–30 suggestions. Similarly, Berk et al. (2017) argue that “referees feel the need to demonstrate their intelligence or industriousness to editors by identifying problems in papers.” Ellison (2002a) also attributes a large part of the increased revision effort to changing norms within the profession. The explanation of the long-term trend is that authors see increasingly demanding referee reports, then become increasingly demanding as referees themselves, and thus the profession produces ever longer papers and ever longer revisions. Hence, to a large extent the excessive revision problems might be due to evolving expectations, not due to a particular institutional feature or rule. If these outcomes are part of a drift in norms, rather than a conscious choice, then perhaps these norms should be revisited.

Potential Solutions

There are numerous potential solutions to this problem, but first and foremost, we as a discipline have to decide whether the job of referees is evaluation or improvement. ⁵ If evaluation, we need to drastically change the expectations of and guidelines for referees. Because de facto referees are doing the improvement task, and perhaps excessively so, as the “anonymous coauthorship” characterization of McAfee (2014) suggests.

Nudges and Appeals

A former editor of the Scandinavian Journal of Economics reminds referees of their more limited role during the referee invitation: “Please give me a quick short review that honestly evaluates the contribution rather than spending a long time to list marginal improvements” (Friberg, 2014). The referee guidelines for the journal Theoretical Economics were also amended to stress that the referee job is not to expand the paper: “The primary purpose of the report is to help the Editor make a decision. If a revision is requested, it should then provide sufficient conditions for publication. The referee should show restraint when asking for new material” (Theoretical Economics, 2023). However, referees are often doing unpaid and complicated work, so editors are already happy someone is willing to accept the referee request. Thus, single editors might find it hard to lecture referees on how they are supposed to review—at least as long as the norms in the profession have not changed. Hence, these small nudges are unlikely to change much if it is only single journals or single editors using them.

No Revisions

A radical solution is to have a peer-review process that either accepts or rejects a manuscript, but does not grant revisions. This saves referees time, because they do not have to think about improvements, and it obviously saves authors time, thus expediting publication. The Economic Inquiry is a journal that offers this referee option to submitting authors. The AER Insights uses this policy as a default (AERI, 2023). The new JoF: Insights and Perspectives will also be a no R&R journal (AFA, 2023). Most other journals have not followed the practice. But “there is overwhelming support among our respondents for this type of model, and across all population subgroups,” as Charness et al. (2022, 3.4.2.12)’s survey among economists finds.

If we think the referee job includes improving the manuscript—and this might be reasonable but too excessive currently—then there are still improvements to be made.

Explicitly Separate Essential and Optional

A small but substantial change would be to require referees to separate “must address for me to recommend acceptance” comments and “suggestion, but I won’t be mad if you don’t follow it” comments. One might say this is already a norm, with the separation into “major” comments and “minor” comments, but that’s not enough. The crucial thing is the explicit promise not to punish the authors if the suggestion is not implemented. Referees must be explicit about this, so that there is no ambiguity whether a suggestion can be safely disregarded. Berk et al. (2017) also describe this measure in a guide to referees, but really it should be a profession-wide standard enforced by journals. Econometrica (2023) recently released official referee guidelines exactly asking for this separation.

Max n Changes Per Referee

To avoid long lists of minor changes requested by referees, the profession could restrict the number of requested changes to, say, 3. This forces the referee to think about the main issues and likely saves time, because minor ones do not have to be spelled out. It also sets an expectation in the profession that referee reports do not have to be long and that referees do not have to impress editors by finding every minor issue (Berk et al., 2017). For some manuscripts, the upper bound of 3 might be binding, but if a manuscript has more than 3 serious issues, then perhaps a rejection recommendation is in order anyway. The Econometrica (2023) guidelines already use a limit of 3 essential changes, but other numbers are possible. More smaller suggestions would be allowed as optional.

Editorial R&R Contracts

If the editor gives an R&R, they have to specify which of the referee comments have to be implemented, and if they are, the manuscript must be accepted (Charness et al., 2022, 3.4.2.15). This proposal reduces the uncertainty for authors, and requires the editors to think upfront which changes are really important, before the authors do all the revision work in vain. Clearly, the commitment has limits, as there are degrees of how satisfactorily the requested changes are implemented. Nevertheless, such an arrangement would be expected to reduce uncertainty and rejection rates after revisions. But, this measure adds workload to editors, so will likely be unpopular with journals.

Decision Without Referees

Frey (2003) suggests that referees no longer advise editors on whether to publish a manuscript. Instead, editors take that decision without referee input, and only in case of acceptance are referees invited to provide suggestions for improvement, which however do not have to be followed. Presumably, this proposal is hard to implement, because editors need referee advice on whether the manuscript has merit, except in the rare cases where it is exactly in their field (e.g., Zilibotti, 2014). Plus, ultimately it means that the acceptance decision is made by a single person, rather than a panel of 2–3, which should make decisions even more noisy (see the noisy outcomes section below). Hence, this might not be the most practical proposal.

At Most One Round of Revisions

Another solution could be to restrict the rounds of revisions to one. Hence, there is acceptance or rejection after at most one revision, which limits revision work. It is not clear if such a change would be welcomed by all, since some referees might not be on board after one revision, but might have come on board after two or three. If authors prefer an acceptance after three revisions to a rejection after one, we might see authors prefer the longer but slightly safer option in the status quo. Still, some journals state they aim to have a definite decision after one revision, but this is typically not a commitment.

Referee Awards

A new idea is to introduce a new form of a referee award, chosen by authors. A typical decision uses two or three referee reports. After the editorial decision, the authors can, but do not have to, choose one of the reports for the “best report award.” Once a year, journals send winning referees a certificate with the number of awards they received. These awards might make the refereeing job more attractive, while hopefully generating more reasonable and useful reports for authors. The award should only be given for the initial decision; otherwise, referees might be tempted to cause more revision rounds. And, the award only makes sense if there is more than one referee. A concern might be that it biases referees to make more positive recommendations to get the award. Also, authors might dislike being put in a position to choose between multiple good reports, especially when the referees know their identity. ⁶ Charness et al. (2022, 3.3.3.3) suggest similar recognition, except by editors rather than authors. ⁷ The median economist in their survey is favorable towards this reform.

Referee Rating System

Authors like referees who recommend acceptance or R&R, and dislike those who recommend rejection. But conditional on the recommendation, there are still better and worse referee reports. Journals could implement a referee rating system, so editors can track referee quality as judged by authors. This would allow editors to avoid the worst referees. These ratings should not be shared with referees; otherwise, it might lead authors to inflate ratings to curry favor. It requires a sufficient sample size to make reasonable referee choice decisions, hence this might only work if these ratings are shared between journals, which is difficult. Plus, journals already struggle to find referees, so might not be able to disregard willing but poor referees.

Problem 2: Strategic Refereeing and Conflicts of Interest

Problem Description

Editors routinely recruit authors of related papers to referee a manuscript. The rationale is that these authors know the literature and the methods best; hence, they make for expert referees. Indeed, if they already know the literature and method well, they should be able to provide a review with the least extra effort, thus making this a seemingly efficient choice. But often referees are also authors of “competing” working papers that try to get published at roughly the same time. They might then exhibit territorial behavior, unduly criticizing the competing manuscripts and recommending rejection for strategic reasons. This for example might be due to a belief that only one paper on that topic can be published in a top journal, and they want that paper to be theirs.

The former editor of the Review of Economic Studies writes: “Sometimes authors want to have exclusive rights to some niches, and they block any entrant. […] There are a number of renowned authors whom I never ask to write a report for this reason” (Zilibotti, 2014). Of course, it takes time for an editor to identify such strategic referees, and by then a good number of manuscripts might have been unfairly rejected.

Currently, I am not aware of serious empirical research into this issue, so more work is needed. But there are anecdotes. Berk et al. (2017), for example, write that an editor caught a referee with a competing paper deliberately delaying the referee report, thus delaying publication of the manuscript. The referee was caught later when they made their own paper public. The editor contacted other editors to get them to boycott the offender’s paper.

A colleague reported the following anecdote:

We submitted our manuscript to a top 5 journal. One referee report summarized our manuscript, then argued it is terrible, has no identification, etc., very harsh criticism. After the rejection, we submitted to a journal below the top 5. One report had the exact same manuscript summary as the report from the previous submission—it was copied and pasted—but now the evaluation was extremely positive. All the criticisms vanished. For the same manuscript! Apparently, someone wanted to make sure we would not get into a top 5 journal.

Such unethical behavior is hard to spot for editors, since it would require the comparison of reports across two different journals. But editors usually have no access to reports from previous submissions.

Potential Solutions

Conflict of Interest Policies for Referees

Many journals—though not all—have conflict of interest policies, which state that referees have to disclose if they have been coauthors, former advisors or students, or family members of any of the authors of the manuscript (e.g., Charness et al., 2022, 3.2.2.3). Other conflicts of interest include referees who have competing manuscripts. While conflict of interest policies might describe and rule out such conflicts, they still rely on the potential referee to disclose them to the editor. Strategic referees might not be so transparent. And even if editors know, they might still be open to having the competing researcher serve as referee, given the difficulty of finding willing referees, in which case it is unclear how they can and do use the (more likely biased) recommendation.

Referee Non-Disclosure Punishments

Punishments for not disclosing conflicts of interest might help, such as getting barred from ever submitting to that journal in the future. After all, undisclosed coauthorships or competing papers will be public at some point, so they might be discovered. If journals can coordinate, banning submission at all participating journals would be an even greater deterrent. This could be easily implementable by big publishers (e.g., Elsevier) or families of journals (e.g., AER, AERI, AEJs). Another punishment might be public condemnation after detection, as is sometimes done with plagiarists. The more it hurts, the more effective it should be.

Referee Disqualification

Given that conflicts of interest are sometimes hard to observe for the editor, it is not easy to deal with a referee that is eager to provide a biased report. Journals outside of economics sometimes offer authors to disqualify a few names as referees, which might prevent the worst offenders from being recruited. But this cannot completely solve the issue, for example, because authors are often not aware who is trying to block them from publishing. And it could be abused by authors to exclude fair referees.

Problem 3: Prestige Bias and Other Discrimination

Problem Description

Do prominent authors or those from top universities get a more favorable peer-review process? Do unknown authors or those from lower ranked institutions get rejected more often purely due to their identity and affiliation? Normally, these questions are hard to assess, since the quality of a manuscript is not independent of author fame and institution: Higher acceptance rates of famous scholars can be fully deserved on the merits and need not result from a bias in their favor. However, there are now at least two field experimental studies which conclude that there is substantial bias against unknown authors or those from lower ranked universities. These studies provide very credible evidence of a bias, because they hold manuscript quality constant and exogenously vary author identity:

Huber et al. (2022) ran a field experiment at a finance journal with a manuscript written by a Nobel Prize winner and a mostly unknown early career researcher. This manuscript was peer-reviewed by hundreds of referees with either the author identity hidden, with only the Nobel Prize winner’s name, or with only the early career researcher’s name. The rejection recommendation rate by the referees increased from 23% for the Nobel laureate to 48% when the author identity was unknown, and increased further to 65% with the early career researcher’s name on the manuscript. Thus, the rejection rate almost triples just based on the author name, for the very same manuscript and the very same institutional affiliation (both authors were at the same institution). The magnitude of the effect is staggering.

Tomkins et al. (2017) ran a field experiment with submissions for a major computer science conference. ⁸ Potential referees were randomly divided into two groups, those that see the author names and affiliations, and those that do not. The same manuscript was reviewed by different referees that know the identity of the authors or not. Tomkins et al. (2017) find that famous authors and top universities get a substantially higher acceptance rate when their identity is known, compared to when it is not known, with acceptance recommendation rates increasing by a factor of 1.6 on average.

It is hard to argue with this experimental evidence. Clearly, there is a status bias in peer-review. The former JEEA editor recognizes the issue and argues that journal incentives favor such biases, because the citing profession has these biases: “There are reasons to believe that the peer-review process is biased in favor of established authors. […] Although individual authors may feel that this is unfair, statistical discrimination may be in the best interests of the journal. The obsession of publishers and associations with objective impact factors reinforces such a bias” (Zilibotti, 2014). In other words, famous authors attract more citations due to their name, so the journal might be willing to accept a manuscript, whereas it would reject the same manuscript from a less famous author. ⁹

One might equally suspect that there is a gender bias against female authors, and possibly racial or nationalist biases. In an early experiment, Blank (1991), manuscripts were randomly assigned to be visible or not to referees at the AER. Male authors’ acceptance rate significantly dropped with blinded identity, whereas female authors’ did not change with blinded identity. ¹⁰ Hence, there is some evidence that double-blind review removed a male advantage (rather than removing discrimination against women). In a non-experimental study, Card et al. (2020) find that papers published by female authors receive significantly more citations than comparable male-authored papers, suggesting a higher bar for women in the peer-review process. But a caveat is that the result relies on heavy parameterization to make studies across genders comparable. In contrast, Forscher et al. (2019) find no or little gender and racial bias in National Institute of Health grant proposals. Aside from the fact that these are not journal manuscripts and the field is not economics, referees also knew that the grant proposals were altered. Hence they might have suspected to be part of an experiment, which invites experimenter demand effects.

Finally, one might suspect that the same manuscript is viewed more favorably if the number of authors is larger: Clearly, with more hands on deck, it must be better! ¹¹ But I am not aware of a convincing field experiment investigating such a “team size bias.”

Potential Solutions

Unfortunately, there are no perfect solutions for this issue.

Double-Blind Review

The solution prior to the mid 90s would have been double-blind peer-review. If the referees do not know the author identity, then they cannot discriminate based on identity. But keeping the author identity secret is almost impossible today with Internet search engines and working papers all over the web. Still, double-blind peer-review might at least partly correct such biases, since some referees might not bother uncovering the author identity. Nevertheless, the AER abandoned double-blind review in 2011 (Goldberg, 2012), citing higher administrative costs and the limited ability of referees to detect conflicts of interest. Since the AER was the only top 5 journal with a double-blind review policy at the time, the decision might also reflect conformism.

Perhaps surprisingly, the median economist has a slightly unfavorable view of double-blind review, except for junior and female economists (who are likely most affected by referee biases), for whom the median is neutral (Charness et al., 2022, Figure 15). In finance, double-blind peer-review is standard at the top 3 journals, in economics single-blind peer-review is standard at the top 5 journals, reflecting a surprising difference in otherwise similar disciplines. ¹² The costs of implementing it are small, ¹³ and it works partly even today in decreasing discrimination (Smirnova et al., 2023). It is hard to see why double-blind reviewing is not the standard in economics, as it is in other disciplines.

Voluntary Double-Blind Review

A voluntary double-blind peer-review, where submitting authors choose between single- and double-blind peer-review, is not promising in terms of incentives, because those benefiting from the bias would not exercise this option. Thus, a separating equilibrium might emerge where getting an anonymous manuscript is the same as receiving an unknown author’s manuscript. Despite this concern, in a field experiment, voluntary anonymization helped increase the acceptance rate of unknown authors slightly (Smirnova et al., 2023), though not enough to offset the large biases estimated in the above field experiments. Still, if voluntary anonymization can help, then mandatory anonymization should help even more.

Problem 4: Noisy Outcomes

Problem Description

At the best journals in economics, the reason for rejection is often not a factual error or objective flaw, but a subjective assessment by referees or the editor that they do not like the manuscript or do not see a sufficient contribution. ¹⁴ Since the law of large numbers does not apply with a panel of 2–3 referees, the outcome—that is, whether the majority recommendation is reject or revise/accept—is extremely noisy. The same paper would get a revision one day while getting rejected the next, because different academics ended up accepting the invitation to referee. Thus, the submission process resembles a lottery. This noise favors topics and methods that are considered general interest or might be a fit at a wide array of journals, and disadvantages less fashionable topics or methods that only have one or two decent journals with the topic within their scope.

But how large is the disagreement among referees? A classical experiment in psychology resubmitted 12 journal articles that had been published in the past 32 months at the same journal (Peters & Ceci, 1982). 3 of these articles were identified by editors or referees as already published, so only 9 went through the full referee process. Of the 18 referee reports thus obtained, 16 recommended rejection. So 8 of the 9 papers were rejected. It would seem that getting published was a matter of luck if the same paper could not make it through twice. Of course, by today’s standards, the sample size of 9 is awfully small.

More modern studies use referee recommendation statistics to compute referee agreement for the same manuscript. This requires multiple referees per paper, which is usually the case with economics journals, but less so with finance journals. Welch (2014) shows that the correlation coefficient between referee recommendations for the same manuscript is 0.25 among six economics journals, and 0.15 at two of the top finance journals. ¹⁵ Agreement tends to be lower at higher ranked journals and at finance journals. Using a structural decomposition model, Welch (2014) argues that a referee-specific signal has a 2/3 weight whereas a common-signal only has a 1/3 weight in explaining referee recommendations. Card and DellaVigna (2020, footnote 20) also find a correlation between referee recommendations of 0.25 for economics journals. Hence, agreement among referees is modest at best.

Potential Solutions

Reusing Prior Reports (Mandatory)

Recruiting more referees would make the outcome less noisy, but is prohibitively costly. An effective approach might be a centralized system that keeps track of all prior reviews across journals (with the possibility of the authors briefly responding to previous reports). ¹⁶ But that has its own issues and is hard to implement. For example, it would require all publishers to share information, and all referees to agree to share their reviews across journals. The latter could be achieved by only allowing academics as referees that agree, but that would make finding suitable referees even harder. Moreover, the manuscript might change after the first submission, thus making earlier reports obsolete. But that could be dealt with via brief author responses (“we solved this concern by doing X in the new version”). Charness et al. (2022, 3.2.4.4) are somewhat negative on the idea, since a single report might punish authors multiple times. ¹⁷

Reusing Prior Reports (Voluntary)

Some journals, such as the Economic Journal (2023), allow authors to voluntarily hand in past referee reports from other journals with the submission, though that might lead to a selective omission of negative past reports. Other journals, such as the AEJs in the AEA family or QE and TE in the Econometric Society family, allow the transfer of past reports from the family’s top journal. This system is a limited version of the above centralized solution where selective omission by the authors is not possible. Editors could forgo inviting new referees by using these past reports, and grant revisions in case of marginal rejections at a superior journal or desk-reject with more information. Editors can always discard the prior reports, so this is an improvement in terms of referee time if it sometimes leads to a decision without new review. Given the voluntary nature on both sides, it is hard to see the harm of allowing the possibility. ¹⁸ One reason editors might be reluctant to use prior reports from other journals, without recruiting new referees, is that these prior referees usually cannot be consulted on a revised manuscript later.

Referee Training

A standard way to increase inter-rater agreement is to provide clear guidelines and training (e.g., Sattler et al., 2015), which is largely missing in economics. Referee guidelines, to the degree they exist, do not actually describe what a good manuscript is, and focus instead on procedural points. Moreover, there is typically no training on writing referee reports in graduate schools. Charness et al. (2022, 3.3.3.2) find that a majority of economists view doctoral training on refereeing favorably. However, voluntary training already exists online, and mandatory training during PhD studies may crowd out more important things—like learning how to do research and then doing it.

Defining Standards of Quality

Possibly, stronger norms and agreement on what methods and approaches are valid would help referee agreement. For example, experimental economics has strong norms that subjects should be paid by performance, should not be lied to, and failure to do either of these is a very likely cause for rejection. In experimental psychology, there is no such agreement on how to run experiments (Hertwig & Ortmann, 2001). But such restrictions on valid approaches and methods, let alone topics—if they can be agreed upon at all—might also make economic research more dogmatic, uniform and less innovative (e.g., Akerlof, 2020), and hence do not seem like a good idea. For a similar reason, Osterloh and Frey (2020) recommend the opposite of reducing noise. They propose a lottery to decide upon acceptance or R&R if referees do not agree, in order to reduce “conservatism bias” that punishes innovative ideas and approaches.

The Role of Editors

The discussion of issues in economics peer-review would not be complete without the role of editors. There is not as much literature or research on it, so the public discussion is not as advanced as it is for the role of referees. Hence, this section gives a first overview of potential issues, but I will not make explicit recommendations based on this early discussion.

Editors have far more power and information (e.g., not blinded even under double-blindness) than referees. Editors are the first line of defense against bad scholarship and decide whether to desk-reject or not, usually without rules, controls, or recourse for authors. While desk-rejections are an essential tool to speed up the review-process, it also invites decision-making based on heuristics, which may generate biases, such as using the author affiliation or publication history for the decision. ¹⁹ If the above field experiments with referees (who often spend hours with the manuscript) find a large prestige bias, then editors who spend only minutes might be even more prone to such biases—especially if a prestige bias is in the interest of an impact-factor maximizing journal. Of course, this does not mean that desk-rejections should be abandoned—there are obvious cases way below the bar that should be rejected quickly—but an argument could be made that double-blindness should apply to editors even more than to referees (e.g., Hassel, 2021). ²⁰

If the manuscript passes the desk-rejection stage, the editor selects referees. Again, the editor has a lot of discretion, and can choose soft or tough referees, those with a competing paper or those that do not, all of which affect the chances of positive reports. As before, without a blinded author identity, editors might subconsciously favor more prestigious authors, male authors, acquaintances, etc. Since even anonymous data on referees is closely guarded, research into referee selection is scarce.

Finally, once referee reports are in, the editor decides whether to grant an R&R or whether to reject. Welch (2014) and Card and DellaVigna (2020) show that editors follow referee recommendations quite closely. But given the previous two stages, editors already had a large impact on the outcome by the time referee reports are in. Another issue at this stage might be missing guidance by the editor in case of conflicting or unreasonable referee comments. Editors could use their authority to limit referee overreach, but appear to do so rarely. Perhaps because editors want to appear grateful to referees, they might shy away from explicitly telling the authors that certain referee suggestions should not be implemented. Moreover, at times, editors bring in new referees after the first revision, adding more uncertainty and possibly work for authors. A colleague writes: “You revise the paper but the editor then brings in another referee who is clearly commissioned to trash the paper and then rejects it on that basis.” While we cannot know what is said between editors and referees that are brought in late, there may be good reasons for it (e.g., another referee became unavailable or a new issue arose, requiring new expertise). Still, facing the lottery of two different sets of referees, given the low agreement between referees, could be perceived as unfair.

There is a relatively large literature on in-group bias by editors, trying to answer whether editors that publish colleagues’ or former students’ work more often is favoritism, better alignment of topics/methods, or justified based on manuscript quality. Colussi (2018) shows that authors publish significantly more in a top 5 journal if the editor is at the same institution. This effect cannot be explained by closer alignment of research fields, and suggests there is an easier way in for authors with social ties to an editor. Lutmar and Reingewertz (2021) find that Harvard and MIT publications in the house journal QJE receive fewer citations than Harvard/MIT publications at other top 5 journals, suggesting a home bias in that journal. ²¹ Bethmann et al. (2023) have very similar findings regarding the QJE in-group bias. Hackman and Moktan (2020) suspect the long editorial terms at house journals could cause in-group bias. Brogaard et al. (2014) also find that authors with the same affiliation as the editor publish significantly more often in that journal, but these publications also have higher citations, indicating a higher quality. Rubin et al. (2023) find an editor home bias in major economics journals, where the share of published authors from a country increases dramatically if the journal editor is from the same country. The published articles from the same countries are less cited, suggesting the increased share is not based on quality. Hence, the evidence is not unanimous, yet the major studies tend to find an in-group bias.

A final concern might be editors publishing in their own journals. They might not be in charge of their own submissions, but their editor-colleagues and possibly friends are. In addition, referees might be worried that their anonymity is threatened, since the submitting editor might nevertheless have editorial access to the review system and might uncover the identities of critical referees. All of these factors suggest that a submitting editor might get a more favorable author experience than an author not editing the journal.

These issues suggest there needs to be a discussion in the profession how much discretion editors should have. More checks and balances might increase review times. On the other hand, some targeted changes might also lead to more meritocratic outcomes. Reform candidates include triple-blind review (editors don’t know author identities), algorithms selecting referees purely based on topic (thus avoiding favoritism in referee selection), or no drafting of new referees after the first revision.

Recommendations for Reform

There is no shortage of ideas. Many solutions to improve peer-review have been suggested before (e.g., Berk et al., 2017; Charness et al., 2022; McAfee, 2014), including somewhat radical proposals (e.g., Azar, 2006; Frey, 2003). What is needed for change is coordinated action, especially because part of the problem are expectations of long papers and extensive revisions (Ellison, 2002a), which are not easily changed by a single journal. The following recommendations are the most promising from the set discussed in the problems section.

The referee guidelines of Econometrica (2023) are great: short, to the point, addressing the referee overreach problem, plus setting expectations that reports do not have to be long, thus likely reducing review times. But this is only one journal. This is not enough to reset expectations profession-wide. We likely need a critical mass of journals to enforce similar referee guidelines; then, these might be established as a new standard. The crucial ingredients of the Econometrica (2023) guidelines are the separation in essential and optional requested changes, and the limit of n essential changes. Any other guidelines with similar provisions might do as well. We should get to an understanding that a referee is not a coauthor, and it is not the referee’s job to rewrite the manuscript.

A system of referee awards for the most useful report is worth a trial. There is an implementation cost by journals, but since they are often struggling to recruit referees, they might be interested in a solution that makes the job more attractive. And if it improves reports in addition, then all the better. I am not aware that this system is used yet, so the outcomes are speculation, but it sounds promising in theory.

There is no perfect solution for prestige bias, or gender and racial biases, as double-blind peer-review can be circumvented. Double-blind review can help a little to reduce bias, and given the small costs of implementation, it should be the standard—as it is in many other disciplines.

In summary, my recommendations are (for detailed discussions, see the problems section):

1. To prevent excessive referee asks, journals should adopt the same referee guidelines which require referees to separate essential and optional changes they request, and limit the essential changes to 3. ²²

2. To make referee reports more useful and to motivate referees, let authors chose one referee per submission for the “best report” award (excluding resubmissions).

3. To minimize strategic refereeing, journals should adopt the same conflict of interest policies, which describe conflicts of interest including competing research, and require referees to disclose these conflicts to the editor. Non-disclosure should be punished, for example, by banning the referee from submitting to all journals using these same policies. Editors should avoid using referees with conflicts of interest.

4. To minimize review biases, double-blind peer-review should be used.

5. To reduce noise in review outcomes and to reduce review times, journals should accept prior referee reports without commitment to use them or to forgo inviting new referees. ²³ If possible, transfer of reports between journals should be facilitated, to avoid selective disclosure of reports by authors. ²⁴

6. To speed up review times, 4 week review deadline should be set and referees should be paid for timely reports (at journals that take a submission fee).

Concluding Remarks

The functioning of the peer-review process is something that affects every publishing economist, and indirectly even the public, because journals give research visibility. Indeed, since hiring and funding decisions often depend on journal publications, the peer-review process by extension determines what kind of research is being done. Many have debated solutions and proposals to improve peer-review, but ultimately journals need to make changes at the source. These need to be coordinated changes, so that we avoid a confusing state with each journal having their own guidelines and rules. This is especially important regarding referee guidelines and conflict of interest policies, which need to be fairly similar and common to shape expectations and norms, so that the issues discussed here are minimized.

Open questions for research remain. How pervasive are conflicts of interest in refereeing? What impacts do they have on publication decisions? Which measures work against them? Do the new Econometrica-style referee guidelines have a measurable impact on referee reports (e.g., length), refereeing times, and extent of revisions? Has the long-term increase in the share of empirical research (Hamermesh, 2013) contributed to the issues in peer-review, and do they affect empirical or theoretical manuscripts more? Such research is important to inform a possible reform effort in economics peer-review further. And more research is needed on the role of editors as well, in particular on desk-rejection decision-making and on referee selection.

This sort of research is not possible without data. Journals need to protect the anonymity of reviewers, but blocking all access to peer-review data also means we cannot diagnose problems and come up with solutions. The research on refereeing so far often relied on influential researchers—often current or former editors—getting exclusive data access (e.g., Welch, 2014). It would be desirable to make such data in anonymized form more widely available.