---------- Forwarded message ---------- Date: Mon, 2 Oct 2000 11:31:33 +0100 From: Stevan Harnad <harnad@COGLIT.ECS.SOTON.AC.UK> Subject: The Number of Times a Scientific Article is Read (fwd) Donald King <dwking@umich.edu>: There are basically 2 ways we (King Research) and others have estimated readership of articles: http://www.cogsci.soton.ac.uk/psyc-bin/newpsy?11.084 The first method involves providing survey respondents with a list of recently published article titles and asking them which articles they have read. While a useful indicator of reading, it underestimates total reading of an article because (1) an article may be read multiple times for various purposes by a scientist; (2) there may be subsequent first readings of articles following the time of observation (usually about 2 months following publication); (3) a substantial amount of article reading comes from sources other than original published issues (e.g., preprints, reprints, interlibrary loan, document delivery, and distributed photocopies); and (4) the article information content is passed on by informal/interpersonal means. The second method involves the combination of 2 estimates: total amount of reading observed from surveys of scientists (i.e., average readings per scientist times total number of scientists) divided by the total number of articles published as observed from a sample of scholarly journals. The latter method involved 2 national surveys of U.S. scientists (under National Science Foundation contracts and OMB approval), the recent surveys (1993-1998) were from self-selected organizations who had asked us to conduct surveys within their organizations (e.g., U. of Tennessee, Johns Hopkins, DuPont, Bristol-Myers Squibb, Procter & Gamble , etc.). These survey methods are described in more detail in our book of the same title (pp. 111-117) and the journal tracking surveys (pp. 119-120). All of the surveys used statistical samples, but have sources of bias. For example, there is no recorded universe of scientists from which to conduct a national sample; thus, we sampled from professional society lists and carefully weighted for overlapping strata. However, the 2 general methods seem to come up with reasonably similar results. Some results using the first method above are as follows. A survey by Garvey and Griffith in the 1960s involving psychology articles yielded estimates of an average of 520 readings per article. Machlup in the late 1970s surveyed readership of 8 recently published economic journals which had 1,240 readings per article. King Research in 1978 sent the table of contents of 37 articles from an issue of the Journal of the National Cancer Institute to a sample of cancer researchers and observed an average of 1,800 readings per article (including projections for future reading based on our reading "decay curves"). In 1977, our national survey of scientists yielded an estimate of 244 million readings by U.S. scientists and 382,300 articles published in U.S. journals, yielding an average of 638 readings per article. The estimated readings took into account the age of articles read, as well as other adjustments for US/non-US authorship and reading. A similar approach in recent years (recognizing that average readings per scientist is not from a national sample) is about 900 readings per article. The difference over time (638 vs 900) reflects an estimated increase in readings per scientist and slight decrease in number of articles published per scientist. Although there is no really perfect way of estimating total readership of articles (including "hits" and "downloads"), the various methods seem to come up with similar results. The same holds true with averages of the amount of readership (and time spent reading) per scientist. I hope that you will keep us informed about the LANL observations. The 300 readings per title per year seem reasonable to me, based on our previous studies of preprint readership. Note, however, that our averages are for the "life" of the article. By the way, I revised our 1977 estimate of amount of readership of article separates after going back to the raw data and combining observations of sources and form of articles read. The revised estimates are: 43 million readings from separate copies of articles (preprints 5 million; reprints 7 mill.; ILL 4 mil.; authors 20 mill.; colleagues 7 mill.). Our recent estimates of over 100 million readings of separate copies still holds. We have surveyed scientists from a National Lab. to compare their information patterns now with mid-1980s observations. We also asked about reading from electronic journals and digital databases including LANL, did they read from the screen or download, how they found out about the preprint, etc. We'll keep you informed. We are also currently surveying scientists at the University of Tennessee as a follow-up to our 1993 survey. All of these surveys rely on observations of a critical incident of reading to establish means of identifying articles, sources of the article (including from separate copies), and consequences of use. I hope this helps. Good luck with your project. Donald W. King Carol Tenopir > ---------- Forwarded message ---------- > Date: Mon, 18 Sep 2000 12:58:29 +0100 (BST) > From: Ian Hickman <ijh198@ecs.soton.ac.uk> > To: tenopir@utk.edu > Cc: Open Citation -- Stevan Harnad <harnad@ecs.soton.ac.uk>, > Les Carr <lac@ecs.soton.ac.uk>, Steve Hitchcock <sh94r@ecs.soton.ac.uk>, > Tim Brody <tdb198@ecs.soton.ac.uk>, Zhuoan Jiao <zj@ecs.soton.ac.uk> > Subject: The Number of Times a Scientific Article is Read > > Dear Dr. Tenopir, > > We are conducting research into the usage of the Los Alamos National > Laboratory Digital Physics Archive (arXiv). > > Based on the download figures for the UK mirror of arXiv, we have roughly > approximated that a paper in the archive gets read about 300 times per > year. > > In your article "Towards Electronic Journals: Realities for Scientists, > Librarians, and Publishers", you state that currently scientific articles > are read about 900 times per year, could you describe the data and analytic > method on the basis of which you arrived at this figure? > > --- > > Ian Hickman, > Department of Electronics and Computer Science, > University of Southampton.
