Jonas Holmström, Research Assistant
The measure for calculating cost per reading (CPR) of journal articles is reviewed, and a way to adapt this measure to articles in open access journals is proposed. The traditional subscription based publishing model is compared with the open access model, and similarities are identified and used when calculating CPR for the two different types of publishing. Challenges with interpreting statistics are discussed as well as the difficulty of estimating the number of readings from the number of downloaded articles. Finally, the potential use and implications of the CPR measure for open access publishers and institutions are discussed.
A recent article by King et al. (2003) provides an overview of library economics and presents a conceptual framework for library economic metrics. This article deals with the cost per article reading (CPR), which is one of the measures mentioned. The cost per article reading has been around for a long time (Tenopir and King, 2000). However, the arrival of electronic journals has given libraries and publishers new tools for measuring article use and subsequently the CPR measure has gained importance.
Anyone remotely familiar with library economics and scholarly journal publishing knows that there is more than one way to calculate the costs and benefits of these products and services. The economics of libraries and scholarly publishing has been subject to numerous studies, which have been made for a number of different reasons. Some extensive studies have reported on the costs and benefits of library provided products and services, e.g., Griffiths and King (1993). Recently the spiralling prices of journal subscriptions have attracted an increasing amount of economic analyses, e.g., Tenopir and King (2000), Bergstrom (2001), Björk et al. (2002), and La Manna (2003) to mention just a few. Journals have been studied from just about every possible angle. Fixed costs and marginal costs have been calculated and then related to the number of subscriptions, articles, pages, citations etc. In a recent review of journal publishing Friedlander and Bessette (2003) found there is surprising variability in the way publishing costs are modeled. Furthermore, Tenopir and King (2000) note that journal pricing is extremely complex, and Quandt (2003) argues that there are two primary reasons why the pricing of journals is complicated. First, publishers practice price discrimination, and second, publishers bundle journals. The wide variety and size of journal publishers and their increasingly complex pricing policies make economic comparisons of the journals publishing system very difficult and subject to a large amount of guessing. However, price is often a function of demand rather than costs.
The rapid increases of journal subscription prices and the possibilities offered by electronic publishing have spurred the development of a new business model for journal publishing known as the open access model. It is difficult for institutions to compare the costs and benefits of subscription-based and open-access-based journals since they charge for their services in totally different ways. This article proposes a method for measuring the cost per article reading of open access articles. This is a new approach since all recent research related to the CPR measure deals with subscription-based journals (Cox 2003, King and Montgomery 2002, Montgomery and King 2002, Plasmeijer 2002, and Scigliano 2002). Since open access journals are available free of charge, this might seem to some like a ridiculous idea . Future discussion will determine how useful the proposed measure is. It is the author's opinion that the suggested model is useful for institutions such as library consortia or universities when they consider whether they should pay article-processing fees to open access publishers or not.
The article reviews previous CPR studies. It then analyses the differences and similarities between subscription-based publishing and open access publishing and presents a method for measuring the CPR for open access journals. Finally, a case study of the CPR of open access articles at the University of Helsinki is presented and conclusions are drawn.
A Review of Previous Studies of Cost per Article Reading
In this article the cost per article reading measure is not a cost to be paid as such as is the case with pay-per-view. The CPR measure is a derived metric in the framework presented by King et al. (2003). The CPR measure relates the cost of resources to the amount of use and is therefore a measure of cost-effectiveness.
Calculating the cost per article reading is the easy part; just divide the cost of providing readings with the number of article readings. The tricky part is to get reliable data on the number of readings and the costs. Tenopir and King (2000) define readings as reading beyond the title and abstract to the body of the article. But, actually measuring readings can be a very difficult. In the print era readings were estimated by measuring re-shelving counts or through surveys that used the critical incident technique to estimate the number of article readings per reader. Re-shelving counts and surveys are expensive and time-consuming ways to collect data. The emergence of electronic journals makes the collection of statistics a lot easier. However, determining what constitutes a reading in the electronic area is not entirely straightforward. The COUNTER project  aims to make usage statistics comparable across publishers and better match actual readings. But we still have a long way to go before the standards are fully developed and implemented. Therefore, the relationships between readings, use and downloads are discussed in the context of each of the reviewed cases below. Determining which costs to allocate to which readings are also difficult due to the way library consortium allocate costs between members.
Montgomery and King (2002) consider cost per use to be the most meaningful measure of a journal's value. The investment bank Morgan Stanley (2002) also notes that libraries are increasingly considering this measure when determining which journals to cancel. Elsevier realises the importance of driving usage of their journals to achieve competitive CPR values and in 2003 released the pamphlet 15 Ways to Promote Effective Use of Online Resources . Below is a review of the most recent CPR research.
The University of Wisconsin-Madison Libraries has calculated cost per use statistics for print journals. A sample of the cost per use data for single journal subscriptions in 1999 is listed in Table 1. These figures are indicators of journal cost per use, not precise measures of readings .
The numbers in Table 1 clearly shows that the cost per use measure alters the perception of value for money.
Scigliano (2002) performed a cost-benefit analysis of the licensing of Annual Reviews Online (AR Online) by 14 members of the Ontario Council of University Libraries (OCUL). AR Online is produced by a non-profit scientific publisher and includes some of the most widely cited titles in 29 subject areas. Since the subscription year had not yet elapsed, use data was estimated based on data from January to March 2002. The cost of the 2002 subscription for OCUL is presented in Table 2no overhead costs are included. Each institution paid the same price.
* The average CPR per institution averaged for all 14 institutions
HighWire Press, which hosts AR Online, specifies article downloads as HMTL and PDF downloads. Sometimes a person may first review the HTML file and then download the PDF file. This should ideally be counted as one reading, but is reported as two downloads. Scigliano reported that after reviewing the conditions under which an article was likely to be counted twice, it was estimated these double counts accounted for about 25% of total downloads. The numbers in Table 2 take this into account.
In two articles, Donald W. King and Carol Hansen Montgomery reported on the costs associated with the migration from a print journal collection to an electronic journal collection at Drexel University (King and Montgomery, 2002 and Montgomery and King, 2002.) Some of the costs and usage statistics from these articles are presented in Table 3. The Drexel project aimed to capture all costs and estimated that total fixed cost for the electronic journal collection equals $0.45/use. This cost is not included in the data in Table 3.
* The CPR has been calculated by subtracting 25% of the uses to get a better approximation of actual readings. For example, the CPR for average of all journals has been calculated as follows: 177,000 uses minus 25% double downloads equals 132,750 readings. The cost of $404,000 is divided by 132,750 readings, which gives the cost of $3.04/reading.
Montgomery and King reported that no publisher- or vendor-supplied use data was available for the individual subscriptions of electronic journals, but made an extrapolation to obtain a "guesstimate". The Drexel studies also reported on cost and use of journals included in full-text databases. These journals are extremely cost efficient on a per use basis ($0.40/use). However, these numbers have not been included here because they are not seen as alternatives to open access journals (or to subscription based journals for that matter).
A readership survey was performed among Drexel faculty and doctoral students (response rate 20 percent and 30 percent, respectively) to complement the usage data provided by vendors and publishers (King and Montgomery, 2002). The survey estimated that faculty together with doctoral students and master's students totaled 198,100 readings from the electronic journal collection. King and Montgomery did not have survey results for undergraduate students, but they estimated that undergraduate students totaled 101,000 readings, based on corresponding library collection use observed by others. This brings the total number of readings to 299,100 from the electronic journal collection. King and Montgomery reported that use data provided by vendors and publishers indicate the electronic collection has an annual use of 400,000. In the OCUL-study, Scigliano (2002) estimated double counts would account for about 25% of downloads. Subtracting 25% from 400,000 uses yields a total of 300,000 readings, whichall things consideredis almost identical to the 299,100 readings reported in the reader survey. The last column in Table 3 presents the average cost per download adjusted for the double download effect.
Cox (2003) calculated the cost per article use for Emerald and the Institute of Physics Publishing (IOPP). He asked the publishers to define full-text access as comprising a download or printing of the articlebrowsing was not included. Both publishers provided aggregated data for all their customers during January 2002 - December 2002. The data is presented in Table 4.
* Cost is actually publishers' revenue attributed to electronic sources. This number has been reverse engineered by multiplying use with cost/use.
The largest portion of publishers' revenues is generated by individual journal subscriptions for both print and electronic access. Cox solved this by making a subjective judgment and assumed that fifty percent of the subscription revenue should be attributed to the electronic version. The numbers in Table 4 are therefore less reliable and useful to the aims of the study presented in this article than they would have been if only data for "pure" e-journal subscriptions had been used.
The Two Publishing Models Revisited
Some might argue that open access publishing and traditional subscription based publishing have totally different business models. This is in some sense true. However, both business models also have similarities that can be used to make economic comparisons. Traditionally, authors have transferred the copyright along with their articles to the publishers. The publishers then hold exclusive rights to the articles and sell articles bundled in journals to subscribers. In some of the open access journals, authors or their institutions pay to have their articles published and made freely available on the Internet. The open access business model differs in two critical aspects from the subscription-based model. First, the cost of the publishing process has been reallocated from subscription fees to publishing fees. In other words, the publisher's customer is no longer the reader but the author. Second, the services for which the publisher charged have changed from access services to publication services. In conclusion, open access and traditional publishers provide different services to different customers.
How then can these two models be compared? By focusing on the things they have in common. Figure 1 presents a stripped down model of what goes on between publishers and institutions. In the subscription-based model, subscription fees enable publishers to provide institutions with access to articles. In the open access model, article processing fees enable publishers to provide institutions with access to articles. Hence, both models essentially function the same wayinstitutions (representing authors and readers) transfer money and manuscripts to publishers and in return gain access to articles.
Some clarifications need to be made relating to Figure 1. First, the institution usually pays authors' article processing fees as well as readers' subscription fees. Second, the author is often, but not always, the same person as the readerthere are those who read but do not write. Third, access to open access articles is free to everyone. However, someone has to pay the publisher to make the articles accessible. So, open access articles are free to access but not free to publish. An illustrative example can be used to prove that open access articles are not free to everyone. According to Tenopir and King (2000) the annual production of articles is about 2 million. If the open access publisher BioMed Central published all these articles at a cost of $500/article  the total annual cost of open access would be $1 billion . This figure might be less than the sum libraries now pay publishers for journal subscriptions, but it is definitely not free and someone has to pay. Fourth, it is possible to be a free rider  in the open access model by accessing the articles without ever paying for them (directly or through one's institution). To use biology terminology, free riders are often considered to be parasitesusing others without contributing anything themselves. In the world of open access, free riders and the institutions/authors who pay article processing fees are in a mutually beneficial relationship since free riders benefit from having free access to articles and in return cite the articles they read, which is beneficial to the paying institutions. One might consider the term "symbiotic free riders" (or perhaps symbiotic free readers) to be an adequate description of those who access open access articles but do not pay article processing fees. It is also worth pointing out that everyone cannot be a free rider . It is important to note that in this article, free riders and the benefits they enjoy are excluded from the CPR measure developed here, since the calculations are made from the point of view of the paying institution.
The analysis above shows that in both models money is exchanged for access to articles. The CPR studies reveal that CPR is calculated by relating money to the amount of readings. There is no reason why one cannot do the same for open access journals. Subsequently, the CPR for open access articles is calculated by relating the article processing fees to the number of readings from the articles provided by the publisher. The next part presents a case study of this approach at the University of Helsinki.
The Cost per Reading of Open Access Articles at the University of Helsinki
It might be worth restating that the CPR measure is a measure of cost-effectiveness and not a pay-per-view fee. This part presents a case study of the CPR measure for open access articles, and the data is from the University of Helsinki and BioMed Central (BMC). In 2002, the University of Helsinki became an institutional member of BMC. The annual membership fee is $6,243  and enables all researchers of the University of Helsinki to publish their articles in BMC journals  without extra payments. For July 2002 - June 2003 there have been 2,126 full-text downloads (HTML and PDF). Downloads average about 177/month and have been relatively constant over the period . The numbers are presented in Table 5.
* Subtracting 25% from 2,216 yields 1,594.5 readings/year at a cost of $3.92/reading.
These statistics cover the BMC site only, and as Jan Velterop noted in a comment on the American scientist forum, BMC-articles are also available from other sources such as PubMedCentral, INIST, Potsdam University, and many self-archives and repositories . Of these other sources, the source most likely to affect the calculations in this article is PubMed/MEDLINE and PubMedCentral . The availability of BMC articles on PubMedCentral might constitute a significant portion of uses by the University of Helsinki not included in the statistics used in Table 5. Furthermore, MEDLINE can be searched through PubMed, which includes links to the articles in the BMC journals. These links are to articles stored on the BMC site as well as to the articles in PubMedCentral. BMC, PubMedCentral, and PubMed/MEDLINE are all displayed on the electronic service collection listing of Terkko . However, Terkko's electronic journal listing more prominently displays links to the BMC site . In conclusion, the readings of BMC articles by the University of Helsinki are most likely higher than the 177/month recorded from the BMC site, so the real CPR is lower.
This article has shown how to calculate the CPR for open access journals. The major advantage of this measure is that it is gives institutions a method to compare the expenditures on subscription-based journals to expenditures on open access journals. The data from CPR analyses can be used as a complement in acquisition discussions and in science policy discussions when institutions consider whether they should sponsor article-processing charges. Institutions should be able to do more accurate calculations using their proprietary data.
The CPR measure presented here is from a reader perspective. Another type of CPR measure can be calculated from the perspective of the author. This is done by relating the article-processing fee to the number of times an author's specific article is downloaded, read or cited. This approach has been used by BMC  but it has one drawback: it is not directly comparable to the CPR measures presented in the literature review above. Both types of CPR measures actually measure different sides of the same coin.
As the data from the University of Helsinki show, major practical problems still remain before CPR can be presented without an accompanying array of footnotes. One of the largest problems in the case study is the lack of statistics from other sites providing access to BMC journals. However, BMC has some statistics, while most other open access publishers may have no statistics whatsoever. If open access publishers find the CPR measure useful, they should consider supplying statistics.
The institutions currently receiving statistics from BMC might want to consider methods to approximate the number of downloads that are now lost. Indeed, two interesting questions for further study would be to ask the institutions how they use the incomplete BMC statistics now and why they requested them in the first place.
Another statistics-related challenge is to separate the number of downloads from the number of readings. The best available estimation of readings seems to be to subtract 25% from the total number of downloads. However, it is likely that the proportion of downloads to readings varies between publishers. Some publishers may have abstracts in HTML and articles in both HTML and PDF, whereas some other publishers may have no abstract and articles in only HTML or PDF. The need for standards and further research in this area is enormous.
The challenges with estimation of cost data are also mounting. In the print era, costs were easy to derive from single journal subscriptions. Today we have consortia deals, bundling of electronic-only journals, and bundling of print and electronic journals, which are increasingly making cost allocation more difficult.
In this article the CPR measure has been seen from the viewpoint of institutions. However, open access publishers might as well use the CPR measure to prove their worth. For example, BMC could calculate the CPR for all articles accessed on its site for all users (including free riders) or only paying institutions.
One very interesting final point that deserves commenting is the basis on which all the CPR measures have been calculated. All measures have related the cost for the current year to the use of all articles, not just those new articles published during the period of agreement. Much of the use comes from new articles, but the larger the back files, the larger proportion of readings from them. This way of measuring CPR benefits older and larger publishers. Publishers have a motive for calculating the CPR for the entire collection and promoting this figure, which is likely to get lower over time as the collection grows. But institutions that have subscribed to a bundle of journals for some time might have reserved the right to use the back files for free. In this case, they might feel that the current way of calculating CPR is inadequate. However, one might also think about electronic publishing as a utility-based service where the publishers provide not just new articles but also access and preservation services to the entire collection.
The author wishes to thank Jukka Englund, Director R&D at the research library of the Medical Faculty in the University of Helsinki for providing cost and usage data for BioMed Central. Special thanks also to Carol Tenopir, professor at the School of Information Sciences at the University of Tennessee and Bo-Christer Björk, professor at the Swedish School of Economics and Business Administration for providing feedback on a draft of the article.
 When I presented this idea to Jan Velterop, publisher at BioMed Central, his first comment was "I'm afraid that your approach is completely the wrong one."
 15 Ways to Promote Effective Use of Online Resources Library Connect: Partnering with the Library Community, Elsevier, 2003. Available at <http://www.elsevier.com/inca/publications/misc/
 For more information about how these numbers have been calculated, see Journal Cost Per Use Statistics University of Wisconsin-Madison Libraries - 1998-1999 at <http://www.wisc.edu/wendt/journals/costben.html> and More Information about Journal Cost Per Use Statistics University of Wisconsin-Madison Libraries at <http://www.wisc.edu/wendt/journals/costben/mcostben.html>.
 "See How much is BioMed Central charging?" an FAQ from Frequently asked questions about BioMed Central's article-processing charges at <http://www.biomedcentral.com/info/about/apcfaq#howmuch>.
 It is possible to calculate the annual total cost an institution would have to pay for access to all articles if all journals were open access. This is done by multiplying the number of articles published annually by an institution with the article-processing charges. This sum can then be compared to the sum an institution pays for journal subscriptions.
 At a minimum, at least one institution has to pay to make at least one article accessible. In a world where every article is published according to the open access model, the only free riders are those who do not publish articles.
 Corresponds to �6.600 converted at the average exchange rate for 2002
 Figures provided to the author by Jukka Englund, Director R&D at Terkko (the research library of the Medical Faculty in the University of Helsinki)
 See "The True Cost of the Essentials (Implementing Peer Review)". Available at <http://www.ecs.soton.ac.uk/
 See page 23 in "What can BioMed Central offer Lund University" by Camille Seecharran. Available at <http://www.lub.lu.se/new_top/omlub/BioMedCentral.pdf>.
[Friedlander and Bessette] Friedlander, Amy and Rändi S. Bessette. The Implications of Information Technology for Scientific Journal Publishing: A Literature Review. NSF 03-323, Project Director, Eileen L. Collins, National Science Foundation, Division of Science Resources Statistics. 2003. <http://www.nsf.gov/sbe/srs/nsf03323/start.htm>.
[Griffiths and King] Griffiths, José-Marie and Donald W. King. Special Libraries: Increasing the Information Edge. Washington (D.C.), Special Libraries Association, 1993. ISBN 0-87111-414-3.
[King, et al.] King, D.W., Peter B. Boyce, Carol Hansen Montgomery, and Carol Tenopir. Library economic metrics: Examples of the comparison of electronic and print journal collections and collection services. Library Trends 51(3), Winter 2003, 376-400
[King and Montgomery] King, D.W. and Carol Hansen Montgomery. After migration to an electronic journal collection: Impact on faculty and doctoral students. D-Lib Magazine 8(12). 2002. <doi:10.1045/december2002-king>.
[Montgomery and King] Montgomery, Carol Hansen and King, Donald W. Comparing library and user related costs of print and electronic journal collections: A first step towards a comprehensive analysis. D-Lib Magazine 8(10), 2002. <doi:10.1045/october2002-montgomery>.
[Morgan Stanley] Morgan Stanley Equity Research. Scientific Publishing: Knowledge is Power. 2002. <http://www.econ.ucsb.edu/~tedb/Journals/morganstanley.pdf>.
[Quandt] Quandt, Richard, E. Scholarly Materials: Paper or Digital? Library Trends 51(3), Winter 2003, 349-375
[Stiglitz] Stiglitz, Joseph, E., Economics of the Public Sector. 2nd ed. New York, 1988. ISBN 0-393-95683-0.
[Tenopir and King] Tenopir, Carol and Donald W. King. Towards Electronic Journals. Washington (D.C.), Special Libraries Association, 2000. ISBN 0-87111-507-7.
Copyright © 2004 Jonas Holmström