Examination of readership in recent years shows that most scientists don't read every article in every issue of a journal, even if they read some articles. Scientists were asked how many articles they read in the last year from the journal most recently read. The results are summarized below. These results show that in approximately 53 percent of the journals a scientist reads, he or she reads five or fewer articles a year, and in 80 percent of the journals, he or she reads ten or fewer articles. Only in about 1 percent of the journals does a scientist read fifty or more articles per year.

Table 7: Number of Articles Read per Journalby Scientists with Personal Subscriptions: 1995
 Number of Articles Read per Journal 1-5 6-10 11-15 16-20 21-25 26-50 51-100 over 100 Percentage of Scientists 53.0% 27.5% 7.0% 5.1% 2.7% 3.8% 0.7% 0.1%
Source: King et al. 1977-98.

Of course, scientists read only some articles more than once, and they are more likely to do so in frequently read journals: 65.5 percent of articles in frequently read journals are read more than once, versus 43.3 percent of the articles in infrequently read journals. However, university scientists report that the usefulness of the articles in infrequently read journals is greater. Average importance ratings (1 = not at all important to 7 = absolutely essential) of readings used for teaching are 5.11 for infrequently read journals and 4.64 for frequently read ones. Those read for research purposes have ratings of 5.23 and 4.88 respectively. Thus, articles in infrequently read journals tend to be more important. However, those two sets have approximately equal purchase value to university scientists since they spend approximately the same amount of time reading an article from an infrequently read journal as from a frequently read journal. That was observed in both the university and the non-university studies.

Table 8: Percentage of Articles Read, by Source: 1993-98
 Readings by University Scientists Readings by Non-University Scientists Source of Articles Ten or Fewer More Than Ten Ten or Fewer More Than Ten Personal Subscription 16.1% 54.1% 15.4% 28.9% Library 71.0% 40.5% 66.0% 51.3% Online — — 6.4% 11.8% Other 12.9% 5.4% 12.3% 8.0%
Source: King et al. 1977-98. Note that proportions differ from Table 4 because some scientists did not respond to the question of how many times a journal was read.

Frequency of readings of journals by individuals is a critical measure, because the amount of reading and the price of the journals have contributed to the decision to subscribe to journals. With high subscription prices, the cost per reading of articles in infrequently read journals becomes prohibitive, yet some journals are relatively inexpensive, for example, those that are priced less than \$500 and are read more than fifty times. For those journals, the cost per reading (at most \$10) is less than the cost of most alternative sources of articles, since the typical cost of reading library-provided articles averages approximately \$11.50 per reading (Tenopir and King 1997). If a scientist reads from seventeen journals on average, and 1 percent of all subscriptions have more than fifty articles that the scientist reads, then nearly one million subscriptions would have at least that level of reading. When the price and cost per reading climb too high, scientists most often use library-provided articles as an alternative to personal journal subscriptions. Thus, as demonstrated in Table 8, most scientists (71.0 percent of those in universities, and 66 percent of the others) rely on libraries to provide the journals in which they read 10 or fewer articles over a year.

Nevertheless, some personal subscriptions are infrequently read, and some frequently read journals are provided by libraries. That is partially because a cost break-even point is dependent on the price of journals, distance to the library, and the availability of journal routing. (Approximately 16 percent of non-university readings are in journals in which the scientist reads many articles, but to which he or she does not have a subscription.) For example, at an average distance to the library and a price of \$100, the break-even point is fourteen readings. In other words, for fewer than fourteen readings it is less expensive to go to the library, while above that number it is less expensive to subscribe. The cost to scientists of subscribing includes both ordering and maintaining journals as well as the time to look up citations or to browse. The cost of using a library is based on reading approximately three articles per visit and includes the time to locate, obtain, and/or photocopy the article. At a journal price of \$250, the break-even point is thirty-three readings; at \$500 it is sixty-five readings; and at \$1,000 it is 128 readings.

Browsing is by far the most prevalent means of identifying articles, but Table 9 shows that only 34 percent of the articles in infrequently read journals are identified by browsing by university scientists, compared to 54 percent of the articles in infrequently read journals identified by other scientists. That suggests that electronic journals should provide a strong browsing capability, regardless of the frequency with which journals are read. Online searching of indexes and abstracts also is important for identifying articles located in infrequently read journals.

Table 9: Percentage of Articles Read, by Means of Identification: 1993-98
 Readings by University Scientists Readings by Non-University Scientists Means of Identifying Articles Ten or Fewer More Than Ten Ten or Fewer More Than Ten Browsing 34.4% 80.0% 54.0% 68.7% Online Searches 18.8% 3.3% 2.7% 9.8% Citations 12.5% 3.3% 11.1% 4.9% Other Persons 12.5% 6.7% 15.8% 9.8% Other 21.9% 6.7% 6.4% 6.8%
Source: King et al. 1977-98. Note that proportions differ slightly from Table 3 because some scientists did not respond to questions of how many times a journal was read.

### Frequency of Reading of Library-Provided Articles

The distribution of reading of library-provided journals tends to be skewed, because many journals are infrequently read while some are extensively read. For example, Kingma (1995) collected data on the use of 787 journals in university libraries. We show his results in Table 10 (adjusted to readings). In about 30 percent of the journals, fewer than twenty-five articles were read during the year, but in 6.6 percent of the journals, more than five hundred articles were read. Like individuals, libraries have the alternative of obtaining article copies from interlibrary borrowing or document-delivery services if the cost per reading of journal subscriptions is too high. In fact, in comparing costs of subscribing and obtaining copies of articles elsewhere, the break-even point is about ten readings when a journal is priced at \$100. The break-even point for a \$250 journal is nineteen readings, thirty-four readings for a \$500 journal, and sixty-four readings for a \$1,000 journal. The cost per reading of heavily read journals — those in which more than five hundred articles are read a year — is very low. The costs of interlibrary borrowing and document delivery range from \$6 to \$27 (Tenopir and King 1997; Kingma 1995; and Jackson 1997) with costs partially dependent on the number of transactions due to economies of scale. If copies of electronic-journal articles can be made available at a much lower cost, the dynamics of journal purchasing for both individuals and libraries will change dramatically.

Table 10: Readings of Library Journals: 1995
 Total Readings per Journal Percentage of Journals Cumulative Percentage 1-10 12.9% 12.9% 11-25 17.1% 30.0% 26-50 17.6% 47.6% 51-100 18.3% 65.9% 101-250 19.1% 85.0% 251-500 8.4% 93.4% over 500 6.6% 100.0%
Source: Tenopir and King 1977.

The use of journal collections in libraries is fairly extensive. A full 97 percent of scientists at the University of Tennessee and The Johns Hopkins University indicate that they have used their university journal collection at some time, averaging fifty-seven uses per scientist per year. That use (which is not necessarily the same as the number of readings) is in line with the estimated number of readings from library-provided articles (101 per scientist per year, including external libraries). A smaller proportion of scientists outside universities indicate that they used their organization's library journal collection (56 percent during 1990-93 and 78 percent during 1994-98). The average number of uses is substantially less: eighteen and thirty-eight uses per year for the same range of years. While the overall use per scientist nearly doubled, the number of uses per user increased from less than thirty-two uses per user per year to forty-nine uses.

We asked journal readers to consider how much it would cost them to find the information they obtained from library-provided articles if there were no library. We asked if they knew about the information prior to reading about it in the article. Could that information — or equally useful information — have been obtained elsewhere? If so, where would they obtain the information (Griffiths and King 1993; and King et al. 1977-1998)? Then, considering the least-expensive alternative source, how much would it cost to obtain the information, including the time to identify, locate, and obtain the source, adding in costs such as telephone calls, transportation, and photocopying.

Approximately 50 percent of the readings involved articles that had information that was new to the reader. Of the readings in which the information was not new, the information could be obtained most frequently from a colleague. Other sources included a journal found at another library (35 percent of articles) and a consultant (7 percent of articles). The estimated cost of alternatives is found to be approximately three times the library's cost to purchase and maintain the journals plus the user's time required to obtain and use them. Thus, shared library collections achieve a substantial savings to users and their employers.

Scientific scholarly journals are read for a long time after publication; the pattern is very much that of a nuclear-decay curve. Whether inside or outside the university, a majority of readings take place in the first year following publication; only 5 percent of the readings involve articles more than fifteen years old. In fact, the oldest critical incident of reading of the sampled scientists involved an article published in 1935.

The age of articles read has implications for electronic journals because most of the older articles are not available in electronic media. When journals become available electronically, they are rarely made available retrospectively (although some digital library initiatives are addressing that concern). The danger is that, in the future, older articles will be ignored because they are not available electronically. The question then becomes whether the older articles are important enough to worry about since they are very infrequently read. Only 15 percent of the articles read by university scientists and 11 percent of the articles read by other scientists are more than five years old.

Table 11: Percentage of Readings by Age of Article: 1993-98
 Age of Articles Read (Years Since Publication) University Scientists Non-University Scientists 1 58.5% 66.9% 2 12.3% 15.1% 3 6.2% 1.7% 4-5 7.7% 5.2% 6-10 9.3% 2.9% 11-15 1.5% 2.9% over 15 4.6% 5.2%
Sources: King et al. 1977-98.

When university faculty said they needed an article for research or teaching purposes, they were asked: "How important is the information contained in this article to achieving your research or teaching objectives?" They rated importance from 1 (not at all important) to 7 (absolutely essential). The average importance rating for research articles less than a year old is 4.78; for those articles over a year old the rating is 5.26. For teaching the averages are 4.76 and 5.00 respectively. The average time spent reading articles less than one year old is 43 minutes per article: it is 61 minutes for articles more than one year old and slightly more for those over five years of age. Outside the universities, the average times are 35 minutes for articles equal to or less than one year old, 49 minutes for those older than one year, and 53 minutes for those over five years old. Our conclusion is that older articles are often read for research purposes; newer articles may be read for casual interest.

In universities, age is clearly a factor in the purposes for which the articles are read. For example, about one-half of the articles under less than a year old are read for teaching purposes, but that proportion declines as the articles become older (30 percent for articles over a year old, and 20 percent more than five years old). The opposite is true for articles read for research purposes. Under one year, 66 percent of the articles are read for research purposes, compared with 74 percent of articles older than one year and 80 percent of those older than five years. That is consistent with the fact that a higher proportion of older articles are to prepare a formal publication such as an article or book. In other words, approximately 26 percent of new articles are said to be read for that purpose, compared with 42 percent of older articles. It seems clear that, although there are fewer readings of older materials, they are useful and valuable to readers and should continue to be made available in the electronic world.

Libraries become the primary source of articles as they become older and the proportion of readings from personal subscriptions decreases dramatically. Nearly 95 percent of readings of articles over five years old are obtained in or by libraries both inside and outside the universities. Some of the readings of older articles obtained from libraries involve articles originally obtained through personal subscriptions that were later discarded. For example, about one-third of older articles have been read at least once before. Some of the earlier articles were initially read to keep abreast of the literature or, perhaps, for teaching purposes, but later a research need for the information arose.

Table 12: Readings by Means of Identification and by Age of Articles: 1993-98
 Age of Articles Read (Years Since Publication) University Scientists Non-University Scientists Source One year or less More than one year More than five years One year or less More than one year More than five years Personal Subscription 55.3% 7.4% 6.7% 29.6% 9.3% — Library 42.1% 70.4% 93.3% 53.0% 72.3% 94.7% Online — — — 3.5% 5.6% — Other 2.6% 22.2% — 13.9% 13.0% 5.3%
Sources: King et al. 1977-98.

The proportion of articles identified by browsing declines substantially over time, as might be expected. Other means, such as automated searches and citations in other articles, become more prevalent as articles age. Thus, it is also important that traditional bibliographic databases and other comparable sources continue to be updated and made available to scientists and librarians in order to help identify older articles in the electronic era.

Table 13: Readings by Means of Identification and by Age of Articles: 1993-98
 Age of Articles Read (Years Since Publication) University Scientists Non-University Scientists Means of Identification One year or less More than one year More than five years One year or less More than one year More than five years Browsing 76.3% 18.5% 6.7% 74.3% 39.6% 15.8% Online Searches 5.3% 14.8% 20.0% 6.2% 17.0% 15.8% Citations 5.3% 11.1% 13.3% 1.8% 18.9% 42.1% Other Persons 5.3% 18.5% 21.1% 8.8% 17.0% 15.8% Other 7.9% 37.0% 38.9% 8.8% 7.5% 10.5%
Sources: King et al. 1977-98.

### Distribution and Use of Separate Copies of Scholarly Articles

We indicated earlier that scientists and their libraries rely heavily on interlibrary borrowing and document delivery to obtain copies of articles not available in their libraries. Scientists also obtain preprints and reprints of articles from authors and publishers. Finally, scientists extensively photocopy articles, particularly those obtained from libraries.

"Ultimately it may be less expensive to obtain copies from electronic sources, particularly if the time the scientists spend locating and reproducing the article is reduced"

The use of interlibrary borrowing and document-delivery services has grown dramatically over the years. Two studies by King Research, Inc. for the U.S. Copyright Office (1978, 1982) showed that there were about four million interlibrary loans (including document delivery) in 1977 and 7.5 million in 1983. It is expected that the number could currently be well over twenty-five million for scholarly scientific articles. A recent survey of 119 libraries by the Association of Research Libraries (Jackson 1997) averaged 12,200 items per library obtained from interlibrary borrowing and document delivery. The 1993 surveys of university scientists showed that nearly 90 percent of them had used library interlibrary loan services, and all scientists averaged 7.5 uses of the service per year. Surveys from 1990 to 1993 showed that about 30 percent of scientists outside the universities used interlibrary loan and/or document delivery services; they averaged 6.1 uses per scientist per year. Studies from 1994 to 1998 showed substantial increases to 64 percent of the scientists being users, with 14.3 uses per scientist per year. If scientists everywhere in the U.S. averaged five items per year, there would be over twenty-five million copies of articles obtained in one year or about four thousand copies per scholarly journal. That, of course, represents a small proportion of total readings (less than 5 percent). However, availability of copies of articles from electronic journals might increase that amount appreciably, particularly if the electronic-access costs are less than the costs of current interlibrary borrowing and document-delivery charges.

In 1977 almost seventy-one million articles were read as separates — articles sold separately from an issue or journal subscription.

Table 14: Separates Received by Scientists: 1977
 Document Type Number Distributed (In Millions) Preprints From authors 2.0 From colleagues 0.1 Reprints From authors 26.0 From publishers 0.6 From colleagues 0.3 From libraries 0.3 Photocopies given to users By ILL 4.0 By scientists themselves 25.0 By library staff 7.5 By colleagues 3.5 By author 1.5 TOTAL 70.8

In 1985 it was estimated from a series of surveys (University Microfilm International 1984) that there were approximately ninety million copies of articles made by or provided to scientists, including photocopies made by libraries from their collections or by scientists themselves (15.5 million and 32.8 million, respectively). Copies also originated from document-delivery services (8.6 million), authors and colleagues (9.7 million), publishers (13.1 million), and other sources such as information analysis centers and clearinghouses (10.3 million). While we have less-definitive survey results concerning the sources of reading for separate copies of articles, there are some indicators that those sources continue to grow at a rate that exceeds the growth in the number of scientists and their reading. Electronic distribution of individual copies of articles could probably replace most of that activity at a reduced cost.

After 1985 we discontinued collecting data on reading from preprints and reprints. One particularly interesting aspect of the distribution of preprints involves Paul Ginsparg's highly successful system for distributing preprints electronically. Initially, the system — developed at Los Alamos National Labs — involved high-energy physics manuscripts, but it has expanded to other physics specialties and other fields of science. Preprints are commonly distributed in physics (Tenopir and King 1997; King and Roderer 1981): in 1981 scientists read some twenty thousand articles from American Institute of Physics journals that were reprints (2,000), preprints (4,500), or photocopies (13,500). At that time physical sciences authors, including chemists, distributed an average of fifty-one reprints per article and approximately 110 preprints per article. Other fields of science generally had fewer preprints but averaged sixty-nine reprints, with mathematics having the fewest (twenty-one) and life sciences the most (110).

Articles in journals held by libraries are photocopied extensively. In The Johns Hopkins University and University of Tennessee libraries, over three-fourths of the scientists say they have used library photocopy machines to copy journal articles, with the amount of use averaging seventeen uses per scientist per year. About 56 percent say they have asked library staff to photocopy an average of six times per scientist per year. Outside of the universities, approximately 62 percent of scientists say they have used the library for photocopying forty-four times per year (thirty times personally and fourteen times by library staff). In companies, at least 27 percent of all photocopies are made of articles in the scientist's own copy of the journal. Colleagues provide approximately 17 percent of all photocopies. Only 2 percent of photocopying involves interlibrary borrowing or document delivery copies. The remainder — 54 percent — is from journals to which the library subscribes. Photocopying is done in order to read articles more conveniently (e.g., while traveling), to permit annotation or highlighting, for retention in personal files for future reference, or to file with laboratory notes or other documentation. Only about one-fourth of those photocopies are loaned or passed on to others (Griffiths and King 1993).

Photocopying library journals costs the library approximately \$1.55 per article in supplies and equipment. The cost climbs to \$14.67 when users' time is included (Griffiths and King 1993). When library staff does the photocopying, the cost to the library is \$2.13 per article as compared to \$4.87 when the scientists' time is included. Thus, photocopying is not inexpensive. In fact, ultimately it may be less expensive to obtain copies from electronic sources, particularly if the time the scientists spend locating and reproducing the article is reduced.

### Serving Small Companies

An appreciable number of scientists work in small high-technology research and development firms. A King Research study in the mid-1980s identified nearly 14,000 such firms with a median of eight employees and an average of approximately 100 employees (Liston et al. 1985). Because of their size, only 13 percent of those firms have information-support staff or librarians, and, because of economies of scale, they spend more than twice the amount per scientist for information that larger companies spend. Higher costs arise from having fewer scientists share in the cost of library collections and from the cost in additional time necessary to obtain information services from academic libraries and other sources. In order to avoid the high out-of-pocket costs of scholarly journals, many scientists in small firms travel to academic, public, and other libraries periodically to read the literature. That, of course, is very expensive in terms of labor costs even though the scientists batch their reading by designating long periods of time in the library to do their reading or to photocopy for later reading. It would seem that electronic journals could be particularly helpful to that community of perhaps 250,000 to 500,000 scientists. That is an area in which journal price differentiation could be particularly effective.