As noted in a few previous messages of mine, I'm interested in some of the features of the Chemistry Preprint Server (CPS, see: http://preprint.chemweb.com/). (But, I have no connection with CPS, and I'm not a chemist).
Yesterday (on May 28) I browsed through the CPS archive. Of the earliest-posted preprints (a total of 32, posted in July or August 2000, i.e. 9-10 months ago), I could identify 14/32 = 44% that have subsequently been published (or, according to the authors, been accepted for publication, or been published in part) in a "brand-name" journal. Of these, 10/32 = 31% could already be found in the ISI Citation Databases. Because of the relatively short interval (9-10 months) between the time when these preprints were posted, and the time when I sought evidence of subsequent publication, these percentages are very likely to be underestimates. I also checked, via the ISI Citation Databases, for one or more publications by any of the authors or co-authors of the 32 preprints. I could identify publications for 28/32 = 87.5%. So (like BMJ's ClinMed NetPrints archive) the CPS preprint archive appears to have been used (at least, initially) mainly by authors who have some previous track record of publication in journals that are included in the ISI Citation Databases. The CPS archive includes a feature that permits visitors to rate the individual preprints on a 1-5 scale. Of the 32 preprints posted 9-10 months ago, 10 have been rated highly (a "4-star" rating; no "5-star" ratings were noted). Of these 10, 6 have already been published, or accepted for publication. This publication rate (60%, so far) is higher than the rate (8/22 = 36%) for the 22 longest-posted preprints that have been rated less highly. But, because the sample size is small, this difference in publication rates isn't statistically significant at the P = 0.05 level (Fisher's Exact Test). At a later time, it will be of some interest to test again the hypothesis that the rating scale may serve a somewhat useful "impact-rating" function (in that the ratings may help readers to find articles that may be more likely to be published in "brand-name" journals). The CPS archive also provides data about the number of views of each individual preprint. Of the 32 longest-posted preprints, 17 received more than 300 views. Of these 17, 9 can be identified as published, or accepted for publication - a publication rate of 53%. Although the publication rate for those that received less than 300 views is 5/15 = 33%, this difference in publication rates is, again, not statistically significant (because of the small sample size). So, again, one may only conclude that the number of views might also serve a useful "impact-rating" function - one that merits further attention at a later date, when a larger sample size of early-posted preprints is available (and especially, preprints that were posted at least a year previously). Of course, it's possible that the 32 longest-posted preprints represent a somewhat biased sample of the entire number of preprints that had been posted to CPS before May 28, 2001 (a total of 226 preprints). Please note that I'm not suggesting that the CPS preprint archive has no flaws. One flaw is (IMHO) that it's a preprint archive (not a "true" eprint archive), in that it appears to be intended for preprints only, not for both preprints and postprints (where, for the postprints, the authors have retained copyright). Although authors can, in the response section of the CPS webpage that provides access to their preprints, post information about the citation for the published version (a "sign-post" function), this "sign-post" capability isn't integrated into the archive as well as it is (for example) at the arXiv archive. Another flaw (again, IMHO) is that it's not on the list of "Open Archives" (at: http://oaisrv.nsdl.cornell.edu/Register/BrowseSites.pl). So, the CPS archive doesn't meet an "inter-operability" criterion. In summary, the CPS archive seems to provide an interesting approach to an "impact-rating" function. It does provide authors with at least some possibility of a "sign-post" function. But, it isn't "inter-operable". (So, it can't easily become part of an (envisioned) universal eprint archive?). Jim Till University of Toronto
