Correction August 1, 2010: the figure of 8 billion from the Ware report, copied from the RIN report, is Great Britain Pounds, not U.S. dollars as I had assumed after reading the Ware report. What this means for my calcuations is that I have underestimated the current STM revenue - by quite a bit. £8 billion, by today's currency conversion, is 12.6 billion USD. The costs of the open access options, however, are accurate. What this means is that I have seriously underestimated the savings for libraries from a full flip to open access. I will recalculate the numbers and prepare a new post in the near future.
Update February 16, 2010: using a more conservative estimate of the PLoS average article processing fee for the last few months, global library cost savings for a full flip to OA at the PLoS average APF for all articles would be $3.1 billion or 56% savings (total cost less than half of current library journal expenditures) (data can be found here). The significance of this more conservative figure is that it factors in the high-end and more expensive PLoS journals, 4 of which are at the top of their fields in impact factor. This illustrates that significant cost savings is compatible with top quality publishing. Since the vast majority of OA journals do not use APFs, actual cost savings from a full flip to OA would likely be much greater.
Based on data supplied by Mark Ware in the recently released report for the Scientific, Technical and Medical Publishers Association (STM) An overview of scientific and scholarly journals publishing, I calculate that library savings from a full flip from subscriptions to open access via article processing fees, at the PLoS One rate of $1,350 would be at least 64%. For the avoidance of doubt, that's about a two-thirds discount. This is presented as an illustration that open access is a wise choice economically, and not just from an access perspective; it is not meant as an endorsement of PLoS One or the article processing fee approach. The majority of OA journals do not charge article processing fees.
In brief, Ware estimates annual STM revenue at $8 billion per year, and quotes Bjork et al on an estimated total peer-reviewed journal article production of 1.5 million articles per year. This is an average of $5,333 revenue for STM for each scholarly article produced in a year. Compare this with the PLoS One article processing fee of $1,350 per article. Factoring in about 70% of STM revenues coming from library sources, the resulting global savings for libraries are 64%. See here for figures.
Other ways of expressing this: PLoS One costs about a fourth of the average revenue per article for STM, or PLoS One is four times as efficient as the average traditional STM journal.
There are many limitations to this brief study. Most of these limitations are reasons why library savings would be greater than 64%. Examples of variables not taken into account:
STM revenue does not take into account non-STM revenue, for journals in the humanities and social sciences and smaller publishers that are not part of STM. The article count, however, is for all disciplines. A higher total revenue would result in a higher average per-article revenue with the current subscription system, which in turn would mean higher library savings with a flip to open access via article processing fees.
This scenario does not take into account non-library revenue for article processing fees, such as authors who can tap into research grant funds for this purpose.
This scenario assumes an average article processing fee of about $1,350 U.S. With an average of $1,535 (BMC standard article processing charge), library savings are still at least 60%. See here for figures.
These figures do not take into account publishers who would like to receive more revenue than standard PLoS One or BMC rates on a routine basis. While libraries would still see savings at article processing fee rates of $3,000 as some publishers are charging - the two-thirds savings scenario comes from supporting high-quality but affordable open access publishers - not just any article processing fee that any publisher might like to charge.
This post is an early sharing of data to be developed for a fuller study in the near future, and part of the Transitioning to Open Access series. Thanks to Andrew Waller for checking some of the math.