If we were to start from scratch today to design a quality-controlled archive
and distribution system for scientific and technical information, it could take
a very different form from what has evolved in the past decade from
pre-existing print infrastructure. Recent technological advances could provide
not only more efficient means of accessing and navigating the information,
but also more cost-effective means of authentication and quality control.
I discuss relevant experiences of the past decade from open electronic
distribution of research materials in physics and related disciplines,
and describe their implications for proposals to improve the implementation of
Keywords: arXiv, peer review, scholarly publishing, scientific publishing
There has been much recent discussion of free access to the on-line scholarly literature. It is argued that this material becomes that much more valuable when freely accessible , and moreover that it is in public policy interests to make the results of publicly funded research freely available as a public good . It is also suggested that this could ultimately lead to a more cost-efficient scholarly publication system. The response of the publishing community has been that their editorial processes provide an essential service to the research community, that these are labor-intensive and hence costly, and that even if delayed, free access could impair their ability to support these operations. (Or, in the case of commercial publishers, reduce revenues to below the profit level necessary to satisfy their shareholders or investors.) Informal surveys (e.g., ) of medium- to large-scale publishing operations suggest a wide range in revenues per article published, from the order of $1000/article to more than $10,000/article. The smaller numbers typically come from non-profit operations that provide a roughly equivalent level of service, and hence are more likely representative of actual cost associated to peer reviewed publication. Even some of these latter operations are more costly than might ultimately be necessary, due to the continued need to support legacy print distribution, but the savings from eliminating print and going to an all-electronic in-house work-flow are estimated for a large non-profit publisher to be at most on the order of 30% . The majority of the expenses are for the non-automatable editorial oversight and production staff: labor expenses that are not only unaffected by the new technology but that also increase faster than the overall inflation rate in developed countries.
A given journal could conceivably reduce its costs by choosing to consider fewer articles, but this would not reduce costs in the system as a whole, presuming the same articles would be considered elsewhere. If a journal instead considers the same number of articles, but publishes fewer by reducing its acceptance rate, this results not only in an increased cost per published article for that journal, but also in an increased cost for the system as a whole, since the rejected articles resubmitted elsewhere will typically generate editorial costs at other journals. Moreover, in this case there is yet an additional hidden cost to the research community, in the form of redundant time spent by referees, time typically neither compensated nor accounted.
One proposal to continue funding the current peer-review editorial system is to move entirely from the subscription model to an "author-subsidy" model, in which authors or their institutions pay for the material, either when submitted or when accepted for publication, and the material is then made freely available to readers. While such a system may prove workable in the long-run, it is difficult to impress upon authors the near-term advantages of moving in that direction. From the institutional standpoint, it would also mean that institutions that produce a disproportionate amount of quality research would pay a greater percentage of the costs. Some could consider this unfair, though in the long-term a fully reformed and less expensive scholarly publication system should nonetheless offer real savings to those institutions, since they already carry the highest costs in the subscription model. Another short-term difficulty with implementing such a system is the global nature of the research enterprise, in which special dispensation might be needed to accommodate researchers in developing countries, operating on lower funding scales. Correcting this problem could entail some form of progressive charging scheme and a proportionate increase in the charges to authors in developed countries, increasing the psychological barrier to moving towards an author-subsidy system. (The other resolution to the problem of unequal resources -- moving editorial operations to developing countries to take advantage of reduced labor costs -- is probably not feasible, though it is conceivable that some of the production could be handled remotely.) A system in which editorial costs are truly compensated equitably would also involve a charge for manuscripts that are rejected (sometimes these require even more editorial time than those accepted), but implementing that is also logistically problematic.
The question for our scholarly research communications infrastructure is: if we were not burdened with the legacy print system and associated methodology, what system would we design for our scholarly communications infrastructure? Do the technological advances of the past decade suggest a new methodology that provides greater utility to the research enterprise at the same or lower cost?
My own experience as a reader, author, and referee in Physics suggests that current peer review methodology in this field strives to fulfill roles for two different timescales: to provide a guide to expert readers (those well-versed in the discipline) in the short-term, and to provide a certification imprimatur for the long-term. But as I'll argue further below, the attempt to perform both functions in one step necessarily falls short on both timescales: too slow for the former, and not stringent enough for the latter. The considerations that follow here apply primarily to those many fields of research publication in which the author, reader, and referee communities essentially coincide. A slightly different discussion would apply for journal publication in which the reader community greatly outnumbers the author community, or vice versa.
Before considering modifications to the current peer review system, it's important to clarify its current role in providing publicity, prestige, and readership to authors. Outsiders to the system are sometimes surprised to learn that peer-reviewed journals do not certify correctness of research results. Their somewhat weaker evaluation is that an article is a) not obviously wrong or incomplete, and b) is potentially of interest to readers in the field. The peer review process is also not designed to detect fraud, or plagiarism, nor a number of associated problems -- those are all left to posterity to correct. In many fields, journal publication dates are also used to stake intellectual property rights (indeed their original defining function ). But since the journals are not truly certifying correctness, alternate forms of public distribution that provide a trustworthy datestamp can equally serve this role.
When faculty members are polled formally or informally regarding peer review, the response is frequently along the lines "Yes, of course, we need it precisely as currently constituted because it provides a quality control system for the literature, signalling important contributions, and hence necessary for deciding job and grant allocations." But this conclusion relies on two very strong implicit assumptions: a) that the necessary signal results directly from the peer review process itself, and b) that the signal in question could only result from this process. The question is not whether we still need to facilitate some form of quality control on the literature; it is instead whether given the emergence of new technology and dissemination methods in the past decade, is the current implementation of peer review still the most effective and efficient means to provide the desired signal?
Appearance in the peer-reviewed journal literature certainly does not provide sufficient signal: otherwise there would be no need to supplement the publication record with detailed letters of recommendation and other measures of importance and influence. On the other hand, the detailed letters and citation analyses would be sufficient for the above purposes, even if applied to a literature that had not undergone that systematic first editorial pass through a peer review system. This exposes one of the hidden assumptions in the above: namely that peer-reviewed publication is a prerequisite to entry into a system that supports archival availability and other functions such as citation analysis. That situation is no longer necessarily the case. (Another historical argument for journal publication is that funding agencies require publication as a tangible result of research progress, but once again there are now alternate distribution mechanisms to make the results available, with other potential supplemental means of measuring impact.)
There is much concern about tampering with a system that has evolved over much of the past century, during which time it has served a variety of essential purposes. But the cat is already out of the bag: alternate electronic archive and distribution systems are already in operation, and others are under development. Moreover, library acquisition budgets are unable to keep pace even with the price increases from the non-profit sector. It is therefore both critical and timely to consider whether modifications of existing methodology can lead to a more functional or less costly system for research communication.
It is also useful to bear in mind that much of the current entrenched methodology is largely a post World War II construct, including both the largescale entry of commercial publishers and the widespread use of peer review for mass production quality control. It is estimated that there are well over $8 billion/year in revenues in STM (Scientific, Technical, and Medical) primary publishing, for somewhere on the order of 1.5-2 million articles published/year. If non-profit operations had the capacity to handle the entirety, and if they could continue to operate in the $500-$1500 revenue per published article range, then with no other change in methodology there might be an immediate 75% savings in the system, releasing well over $5 billion globally. (While it is not likely that institutions supporting the current scholarly communications system would suddenly opt to reduce their overhead rates, at least their rate of increase might be slowed for a while, as the surplus is absorbed to support other necessary functions.) The commercial publishers stepped in to fulfill an essential role during the post World War II period, precisely because the non-profits did not have the requisite capacity to handle the dramatic increase in STM publishing with then-available technology. An altered methodology based on the electronic communications networks that evolved through the 1990's could prove better scalable to larger capacity. In this case, the technology of the 21st century would allow the traditional players from a century ago, namely the professional societies and institutional libraries, to return to their dominant role in support of the research enterprise.
The arXiv  is an automated distribution system for research articles, without the editorial operations associated to peer review. As a pure dissemination system, i.e., without peer review, it operates at a factor of 100 to 1000 times lower in cost than a conventionally peer-reviewed system . This is the real lesson of the move to electronic formats and distribution: not that everything should somehow be free, but that with many of the production tasks automatable or off-loadable to the authors, the editorial costs will then dominate the costs of an unreviewed distribution system by many orders of magnitude. This is the subtle difference from the paper system, in which the expenses directly associated to print production and distribution were roughly the same order of magnitude as the editorial costs. When the two were comparable in cost, it wasn't as essential to ask whether the production and dissemination system should be decoupled from the intellectual authentication system. Now that the former may be feasible at a cost of less than 1% of the latter, the unavoidable question is whether the utility provided by the latter, in its naive extrapolation to electronic form, continues to justify the associated time and expense. Since many communities rely in an essential way on the structuring of the literature provided by the editorial process, a first related question is whether some hybrid methodology might provide all of the benefits of the current system, but for a cost somewhere in between the greater than $1000/article cost of current editorial methodology and the less than $10/article cost of a pure distribution system. A second question is whether a hybrid methodology might also be better optimized for the differing needs, on differing timescales, of expert readers on the one hand and neophytes on the other.
The arXiv was initiated in 1991, before any physics journals were on-line. Its original intent was not to supplant journals, but to provide equal and uniform global access to prepublication materials (originally it was only to have had a three month retention time). Due to the multi-year period from '91 until established journals did come on-line en masse, the arXiv de facto took on a much larger role, by providing the unique on-line platform for near-term (5-10 yr) "archival" access. Electronic offerings have of course become commonplace since the early 1990's: many publishers now put new material on-line in e-first mode, and the searchability, internal reference linking, and viewable formats they provide are at least as good as those of the automated arXiv. These conventional publishers are also set up to provide superior services wherever manual oversight, at additional cost, can improve on the author's product: e.g., correcting bibliographic errors and standardizing the front- and back-matter for automated harvesting. (Some of these costs may ultimately decline or disappear, however, with a more standardized "next-generation" document format, and improved authoring tools to produce it -- developments from which automated distribution systems will benefit equally.)
We can now consider the current roles of the arXiv and of the on-line physics journals and assess their overlap. Primarily, the arXiv provides instant pre-review dissemination, aggregated on a field-wide basis, a breadth far beyond the capacity of any one journal. The journals augment this with some measure of authentication of authors (they are who they claim to be), and a certain amount of quality control of the research content. This latter, as mentioned, provides at least the minimum certification of "not obviously incorrect, not obviously uninteresting"; and in many cases provides more than that, e.g., those journals known to have higher selectivity convey an additional measure of short-term prestige. Both the arXiv and the journals provide access to past materials; and one could argue that arXiv benefits in this regard from the post facto certification functions provided by the journals. It is occasionally argued that organized journals may be able to provide a greater degree of long-term archival stability, both in aggregate and for individual items, though looking a century or more into the future this is really difficult to project one way or another.
With conventional overlapping journals having made so much on-line progress, does there remain a continued role for the arXiv, or is it on the verge of obsolescence? Informal polls of physicists suggest that it remains unthinkable to discontinue the resource, that it would simply have to be reinvented because it plays some essential role not fulfilled by any other. Hard statistics substantiate this: over 20 million full text downloads during calendar year '02, on average the full text of each submission downloaded over 300 times in the 7 years from '96-'02, and some downloaded in the tens of thousands of times. The usage is significantly higher than comparable on-line journals in the field, and, most importantly, the access numbers have accelerated upwards as the conventional journals have come on-line over the past seven years. This is not to suggest, however, that physicist users are in favor of rapid discontinuation of the conventional journal system either.
What then is so essential about the arXiv to its users? The immediate answer is "Well, it's obvious. It gives instant communication, without having to wait a few months for the peer review process." Does that mean that one should then remove items after some fixed time period? The answer is still "No, it remains incredibly useful as a comprehensive archival aggregator," i.e., a place where for certain fields instead of reading any particular journal, or set of journals, one can browse or search and be certain that the relevant article is there, and if it's not there it's because it doesn't exist. (This latter archival usage is the more problematic with respect to the refereed journals, since the free availability could undercut the subscription-based financial models -- presuming the author-provided version is functionally indistinguishable from the later journal version).
It has been remarked  that physicists use the arXiv site and do not appear concerned that the papers on it are not refereed. The vast majority of submissions are nonetheless submitted in parallel to conventional journals (at no "cost" to the author), and those that aren't are most frequently items such as theses or contributions to conference proceedings that nonetheless have undergone some effective form of review. Moreover, the site has never been a random UseNet newsgroup-like free-for-all. From the outset, a variety of heuristic screening mechanisms have been in place to ensure insofar as possible that submissions are at least of refereeable quality. That means they satisfy the minimal criterion that they would not be peremptorily rejected by any competent journal editor as nutty, offensive, or otherwise manifestly inappropriate, and would instead at least in principle be suitable for review (i.e., without the risk of alienating or wasting the time of a referee, that essential unaccounted resource). These mechanisms are an important -- if not essential -- component of why readers find the site so useful: though the most recently submitted articles have not yet necessarily undergone formal review, the vast majority of the articles can, would, or do eventually satisfy editorial requirements somewhere. Virtually all are in that grey area of decidability, and virtually none are entirely useless to active physicists. That is probably why expert arXiv readers are eager and willing to navigate the raw deposited material, and greatly value the accelerated availability over the filtering and refinement provided by the journal editorial processes (even as little as a few months later).
The idea of using prior electronic distribution to augment the referee process goes back at least to . Proposals along the lines of decoupling peer review from arXiv distribution can be found in , and the notion of "overlay" journals is further discussed in ,. A review of various "decoupling" and "author subsidy" models proposed in the mid to late 1990's, taking advantage of new technology to implement improvements in research communication, can be found in . (Note, in particular, the "eprint moderator model" , intended to reduce costs by reducing the amount of material distributed in a commercial manner.) Recent experience in physics and related disciplines continues to reinforce the desirability of experimentation within this model space, with the expectation that similar implementations will prove feasible in other disciplines.
According to the observations above, the role of refereeing may be over-applied at present, insofar as it puts all submissions above the minimal criterion through the same uniform filter. The observed behavior of expert readers indicates that they don't value that extra level of filtering above their preference for instant availability of material "of refereeable quality." Non-expert readers typically don't need the availability on the timescale of a few months, but do eventually need a much higher level of selective filtering than is provided on the short timescale. Expert readers as well could benefit on a longer timescale (say a year or longer) from more stringent selection criteria, for the simple reason that the literature of the past decade is always much larger than the "instantaneous" literature. More stringent criteria on the longer timescale would also aid significantly in the job and grant evaluation functions, for which signal on the year or more timescale remains sufficiently timely. More stringent evaluation could potentially play a far greater role than peer-reviewed publication currently does, as compared to external letters and citation analyses.
Can these considerations be translated into either a more functional or more cost-effective peer review system? As already discussed, editorial costs cannot be reduced by adopting a lower acceptance rate on some longer timescale, but with the same number of submissions considered as currently, and by the current methodology. Instead the simplest proposal is a two-tier system, in which on a first pass only some cursory examination or other pro forma certification is given for acceptance into a standard tier. This could be minimally labor-intensive, perhaps relying primarily on an automated check of author institutional affiliation, prior publication record, research grant status, or other related background; and involve human labor primarily to adjudicate incomplete or ambiguous results of an automated pass.
Then at some later point (which could vary from article to article, perhaps with no time limit), a much smaller set of articles would be selected for the full peer review process. The initial selection criteria for this smaller set could be any of a variety of impact measures, to be determined, and based explicitly on their prior widespread and systematic availability and citability: e.g., reader nomination or rating, citation impact, usage statistics, editorial selection, ... . The instructions to expert reviewers would be similar to those now, based on quality, originality, and significance of research, degree of pedagogy (for review articles), and so on. The objective would be greater efficiency by focusing the comprehensive process not only on a smaller subset, but also that with a higher likely acceptance rate. These are the articles most likely to be archivally useful, and hence merit the enhanced editorial treatment for upgrade into the upper tier, including, for example, text clarifications and other improvements. This would also reduce the inefficient expenditure of community intellectual resources on articles that may not prove as useful in the long-term. Upper tier enhancements could include anything from a thorough blind refereeing to open professional annotation and comment. The upper tier could also combine commentary on many related papers at once. The point is that it's possible to provide more signal of various sorts to users on a smaller subset of articles, without worry about fairness issues of limited dissemination for the rest, and this can be done at lower overall cost than the current system, both in time spent by editors and in elective time spent by referees.
The standard tier would provide a rapid distribution system only marginally less elite than much of the current publication system, and enormously useful to readers and authors. Articles needn't be removed from the standard tier, and could persist indefinitely in useful form (just as currently in the arXiv), available via search interfaces and for archival citation -- in particular, they would remain no less useful than had they received some cursory glance from a referee. Rapid availability would also be useful for fields in which the time to publication time is perceived to be too long. The standard tier availability could also be used to collect confidential commentary from interested readers so that eventual referees would have access to a wealth of currently inaccessible information held by the community, and help to avoid duplication of effort. In addition, articles that garner little attention at first, or are rejected due to overly restrictive policies, only to be properly appreciated many years later, would not be lost in the short-term, and could receive better long-term treatment in this sort of system. Various gradations, e.g., appearance in conference proceedings, would also automatically appear in the standard tier and provide additional short-term signal occasionally useful to non-expert readers.
The precise criteria for entry into the standard tier would depend on its architecture. Adaptable criteria could apply if it is some federation of institutionally and disciplinarily held repositories. The institutional repositories could rely on some form of internal endorsement, while the disciplinary aggregates could rely either on affiliation or on prior established credentials ("career review"  as opposed to "peer review"). Alternate entry paths for new participants, such as referrals from prior credentialed participants or direct appeal for cursory editorial evaluation (not full-fledged peer review), would also be possible. The essential idea is to facilitate communication within the recognized research community, without excessive noise from the exterior . While multiple logically independent (though potentially overlapping ) upper tiers could naturally evolve, only a single globally held standard tier is strictly necessary, with of course any necessary redundancy for full archival stability. Suitable licensing procedures or copyright retention  to facilitate such a system are consistent with the spirit of copyright law, "To promote the Progress of Science and useful Arts" (for a recent discussion, see ).
At the second stage, it might also be feasible and appropriate for the referees and/or editor to attach some associated text explaining the context of the article and the reasons for its importance. Expert opinion could be used not only to guide readers to the important papers on the subject, but also guide readers through them. This would be a generalization of review volumes, potentially more timely and more comprehensive. It could include both suggested linked paths through the literature to aid in understanding an article, and could also include links to subsequent major works and trends to which an article later contributed. This latter citation tree could be frozen at the time of the refereeing of the article, or could be maintained retroactively for the benefit of future readers. Such an overlay guide to the "primary" literature could ultimately be the most important publication function provided by professional societies. It might also provide the basis of the future financial model for the second stage review process, possibly a combination of subscription (electronic, or even print if desired) and upper tier author subsidy. It could subsidize part of the cost of the less selective "peer reviewable" designations in the first stage for those lacking institutional credentials, perhaps together with a first stage "editorial fee" far smaller than for the later full editorial process.
As just one partial existence proof for elements of this system, consider for example the Mathematical Reviews, published by the American Mathematical Society. It provides a comprehensive set of reviews of the entire mathematical literature and an invaluable resource to mathematicians. It currently considers on the order of 100,000 articles per year, and chooses to review approximately 55,000 of those, at a rough overall effective editorial cost of under $140 per review . The expenses include a nominal payment to reviewers, and also curation and maintenance of historical bibliographic data for the discipline. (Mathematician Kuperberg has also commented that "Math Reviews and Zentralblatt are inherently more useful forms of peer review" , though observes ironically that their publishers do not share this conviction.) Mathematical Reviews uses as its information feed a canonical set of conventional mathematics journals. In the future, such an operation could conceivably use some equally canonicalized cross-section taken from a standard tier of federated institutional and disciplinary repositories, containing material certified to be "of peer reviewable quality." While not all upper tier systems need to aspire to such disciplinary comprehensivity, this does provide an indication that they can operate usefully at an order of magnitude lower cost than conventionally peer reviewed journals.
The modifications described here are intended as a starting point for discussion of how recent technological advances could be used to improve the implementation of peer review. They are not intended to be revolutionary, but sometimes a small adjustment, with seemingly limited conceptual content, can have an enormous effect. In addition, these modifications could be undertaken incrementally, with the upper tier created as an overlay on the current publication base, working in parallel with the current system. Nothing would be jeopardized, and any new system could undergo a detailed efficacy assessment that many current implementations of peer review have either evaded or failed.
Acknowledgements: I thank David Mermin, Jean-Claude Guédon, Greg Kuperberg, Andrew Odlyzko, and Paul Houle for comments. This text evolved from discussions originally with an American Physical Society publications oversight subcommittee on peer review, on which I served in early 2002 along with Beverly Berger, Mark Riley and Katepalli Sreenivasan.
 R. Stephen Berry, "Is electronic publishing being used in the best interests of science? The scientist's view", Electronic Publishing in Science II, UNESCO HQ, Paris, 2001 (eds. Sir R. Elliot and D. Shaw), http://users.ox.ac.uk/~icsuinfo/berryfin.htm
 Steven Bachrach, R. Stephen Berry, Martin Blume, Thomas von Foerster, Alexander Fowler, Paul Ginsparg, Stephen Heller, Neil Kestner, Andrew Odlyzko, Ann Okerson, Ron Wigington, Anne Moffat, ``Who should own scientific papers?'' Science 281: 1459-1460 (1998) http://www.sciencemag.org/cgi/content/full/281/5382/1459
 P. Ginsparg "Creating a Global Knowledge Network", in Electronic Publishing in Science II, proceedings of joint ICSU Press/UNESCO conference, Paris, 2001 (eds. Sir R. Elliot and D. Shaw), http://users.ox.ac.uk/~icsuinfo/ginspargfin.htm (copy at http://arXiv.org/blurb/pg01unesco.html)
 This estimate is for the American Physical Society, which publishes over 14,000 articles per year, and derives from figures discussed with its publications oversight committee. The percentage estimated for other publishing operations will vary, especially when editorial time and overhead is differentially accounted. In the discussion that follows, however, it matters only that there will be no windfall savings to publishers from going all-electronic, while employing the same overall labor-intensive methodology.
 Jean-Claude Guédon, "In Oldenburg's Long Shadow: Librarians, Research Scientists, Publishers, and the Control of Scientific Publishing", Proc. ARL Membership Meeting, May 2001. http://www.arl.org/arl/proceedings/138/guedon.html
 See http://arXiv.org/ . For general background, see P. Ginsparg "Winners and losers in the global research village", in Electronic Publishing in Science I, proceedings of joint ICSU Press/UNESCO conference, Paris, 1996 (eds. Sir R. Elliot and D. Shaw), http://users.ox.ac.uk/~icsuinfo/ginsparg.htm (copy at http://arXiv.org/blurb/pg96unesco.html)
 "Brinkman Outlines Priorities, Challenges for APS in 2002", APS News, January 2002. http://www.aps.org/apsnews/0102/010208.html
 S. Rogers and C. Hurt, "How Scholarly Communication Should Work in the 21st Century", The Chronicle of Higher Education, October 18, A56 (1989).
 P. Ginsparg, "First Steps Towards Electronic
Research Communication," Computers in Physics, Vol.8, No.4, Jul/Aug 1994,
see also P. Ginsparg, "After Dinner Remarks", http://arXiv.org/blurb/pg14Oct94.html, presented at the APS e-print Workshop at LANL, 14-15 Oct 1994, http://publish.aps.org/EPRINT/KATHD/toc.html
 P. Ginsparg, "Los Alamos XXX", November 1996 APS News Online, http://www.aps.org/apsnews/1196/11718.html (copy at http://arXiv.org/blurb/sep96news.html)
 Steven Gass, "Transforming Scientific Communication for the 21st Century," Science and Technology Libraries, vol. 19, no. 3/4, 2001, pages 3-18
 David Stern, "eprint Moderator Model", http://www.library.yale.edu/scilib/modmodexplain.html (version dated Jan 25, 1999)
 Rob Kling, Lisa Spector, Geoff McKim, "Locally Controlled Scholarly Publishing via the Internet: The Guild Model", The Journal of Electronic Publishing, August, 2002. http://www.press.umich.edu/jep/08-01/kling.html
John Willinsky, "Copyright Contradictions in Scholarly Publishing", First Monday, volume 7, number 11 (November 2002), http://firstmonday.org/issues/issue7_11/willinsky/index.html
 Private communications from past and current Mathematical Reviews editors Keith Dennis and Jane Kister, based on publicly available data.
 Greg Kuperberg, "Scholarly mathematical communication at a crossroads",
Nieuw Arch. Wisk. (5) 3 (2002), no. 3, 262-264.