Knowledge of the cost-per-use structure of a serials collection—or, at least, knowledge of the effect of one's actions on this structure—is essential in making any cancellation decisions that would result in a transition from ownership to access through document delivery. In general, researchers have demonstrated that whereas expensive journals with little in-house use are better accessed through document delivery, journals frequently used by patrons are best bought through subscription (Feguson and Kehoe 1993; Chrzastowski and Anthes 1995; Gossen and Irving 1995; Kingma and Irving 1996). For example, Gossen and Irving (1995, 49) estimated that if the University of Albany had switched totally to document delivery in 1992, it would have cost $2,900,456 to provide the same access to journal literature that was provided by $1,273,531 in annual subscription costs. To make judgments in these matters, some sort of benchmark is needed.
In a study jointly sponsored by ARL and the Research Libraries Group, Roche (1993) attempted to provide such a benchmark through an analysis of the costs involved in interlibrary loan transactions during 1991. Roche estimated that a research library spends an average of $18.62 to borrow a document. However, examination of the bases of this estimate led to the conclusion that it did not pertain to the question under consideration because it did not take into account the actual cost of the document itself. The estimate only related to the cost of processing the document such as expenses for staff, network and communications, delivery, photocopy, supplies, equipment and software, etc. These types of expenses appear to be more than counterbalanced by the expenses of owning a serial other than its subscription price. The latter expenses include such items as technical processing and binding. In a study of science and mathematics serials, Kingma and Irving (1996, 1–2, 35–38) estimated that the annual cost of owning a title other than its subscription price averaged $62.96, with an additional cost of $0.07 for every use.
Given these facts and the extreme variation of the serials subscription prices by discipline, it was decided that the most reasonable way to use our cost-per-use figures derived by the accounting method was not to compare them against some universal standard but only against each other within defined subject sets. However, for general guidance, two figures can be used. First, it cost the UIUC Chemistry Library an average of $16.76 per document during its experiment with the Chemical Abstracts Service (CAS) from October 15, 1993 to April 30, 1994 (Chrzastowski and Anthes 1995, 145). Second, the mean cost of the documents delivered to the LSU Department of Chemistry through UnCover from the start of the service in October 1, 1995 through June 30, 1997 was calculated. It was virtually identical—$16.92. Only the mean document cost-per-use of the U.S. association titles at the UIUC Chemistry Library was beneath these guidance figures, raising serious doubts as to whether the CAS and UnCover would be able to maintain their pricing structures if heavy cancellations of serials by libraries forced publishers to rely on the sale of individual documents rather than subscriptions for their revenue.
Before testing for the effect of employing the Evaluator's second algorithm with the sociometric measures of ST value, it was considered necessary to gauge the strengh of the relationship of these sociometric measures of value to the accounting measure of value, estimated annual use. Again, the method of analysis was to compute the Pearson product-moment correlation coefficient between the variables in question, then regress one variable on the other—this time the sociometric measure on the accounting measure—to determine the outliers, and then recompute the Pearson product-moment correlation with the outliers excluded. The results were extremely encouraging.
With LSU faculty ratings as the sociometric measure, the initial correlation was 0.72. Three outliers were found, of which two shared the same characteristics with the three outliers found in the correlation of LSU faculty ratings with total UIUC use, i.e., narrow subject focus, lower end of the use distribution, and actual use lower than predicted use. As a matter of fact, one of the titles appeared as an outlier in both correlations. The third outlier was different in that the actual use was much higher than the predicted use, and all could have been the result of either different subject interests at LSU and UIUC or the shortness of the use sampling period. Their exclusion raised the correlation of LSU faculty ratings with estimated annual UIUC use to 0.74—almost the same as with total UIUC use.
Similar results were obtained when total ISI citations were used as the sociometric measure. Here the initial correlation was 0.76, and regression of total ISI citations on estimated annual UIUC use revealed four outliers. One of these had the characteristics of narrow subject focus, lower end of UIUC use, and actual use lower than predicted use, and it had appeared as the same type of outlier in the correlation of total ISI citations with total UIUC use. The other three outliers had an actual use much higher than predicted use, and once again—as with total UIUC use—Chemical and Engineering News appeared in this role. As for the other two outliers of the latter type, they had a relatively narrow subject focus and could have been the result of differing subject interests at UIUC or the shortness of the use sampling period. With the exclusion of the outliers, the correlation of total ISI citations with estimated annual UIUC use rose to 0.82—again virtually the same as with total UIUC use.
The correlation of the sociometric measure of total UIUC use with the accounting measure of estimated annual UIUC use can be dealt with briefly. Initially the correlation was 0.97, and the exclusion of four outliers raised it to 0.98, or virtual unity.
The lesson of these experiments was clear: the accounting measure of estimated annual UIUC use was practically the same as the sociometric measure of total UIUC use and interacted with the other sociometric measures of LSU faculty ratings and total ISI citations in virtually the same way. A plot of estimated annual UIUC use against price showed the same bifurcated pattern, with scientific value concentrating on U.S. association serials and costs concentrating in the serials of the commercial, largely foreign, publishers.
Tests of the effects of employing the Evaluator's second algorithm with all three sociometric measures of scientific value were made on the set of the 120 titles common to the 1993 UIUC use study and SRP pilot project with the LSU Department of Chemistry. In these tests the SAS statistical program was utilized to simulate Evaluator runs with the second algorithm to avoid the complex task of loading the test set into the Evaluator and to keep open the option of exploring other algorithms. The simulated runs were done with the Evaluator default settings of trying to reduce costs by 75%—in this case approximately $120,000—while trying to retain 75% of the scientific value of the serials collection. These default settings had been selected because long experience of working with informetric distributions had indicated them as approximately the maximum optimal levels. The result of each run was similar to the others.
Using LSU faculty score, 27 (22.5%) of the titles costing $59,745.99 (37.3%) of the total costs were recommended for cancellation or nonpurchase. The loss in the sociometric measure of faculty score was 11.0% for a favorable ratio of percentage cost reduction (37.3%) to percentage scientific value loss (11.0%) of 3.4 to 1. With respect to the accounting measure of estimated annual UIUC use, the loss in scientific value was 8.4% for a favorable ratio of 4.4. to 1 in percentage cost reduction to value loss. With total ISI citations as the sociometric measure of scientific value, 30 (25.0%) of the titles with $64,928.20 (40.5%) of the total cost were recommended for cancellation or nonpurchase. The loss in value measured by total ISI citations was 9.6% for a favorable ratio of 4.2 to 1, and the loss in estimated annual use was 9.8%, yielding a favorable ratio of 4.1 to 1. Employing total UIUC use as the sociometric measure of scientific value, 35 (29.2%) of the titles with $77,657.05 (48.5%) of the total cost were recommended for cancellation or nonpurchase. The loss in sociometric value measured by total UIUC use was 11.5%, resulting in a favorable ratio of 4.2 to 1, and the loss in estimated annual UIUC use was 12.1%, giving a favorable ratio of 4.0 to 1. In all cases, the sociometric results were similar to the accounting results in terms of cost reduction to value loss.
Much more importantly, the cost-per-use statistics of all three subsets of serials recommended for cancellation or nonpurchase were much higher than the equivalent statistics for the complete set of 120 serials, which had a mean title cost-per-use of $247.49 and a mean document cost-per-use of $41.30. Concerning the first measure, the mean title cost-per-use statistics of the subsets recommended for cancellation or nonpurchase were as follows for the three sociometric measures: LSU faculty score, $591.61; total ISI citations, $544.36; and total UIUC use, $484.59. With respect to mean document cost-per-use, the statistics were the following: LSU faculty score, $182.99; total ISI citations, $170.24; and total UIUC use, $164.95. The document means were all approximately 10 times higher than the average of $16.76 that it cost the UIUC Chemistry Library for the delivery of a document during its experiment with the Chemical Abstracts Service from October 15, 1993 to April 30, 1994, as well as the average of $16.92 that LSU Libraries paid for documents delivered by UnCover to the LSU Department of Chemistry during the period of October 1, 1995 to June 30, 1997.
An examination of the titles recommended for cancellation or nonpurchase reveals the dominant role of commercial publishers in the cost of ST serials. Of the 27 titles recommended for such treatment based on LSU faculty ratings, 20 (74.1%) were those of foreign commercial publishers, and 6 (22.2%) belonged to U.S. commercial publishers. With respect to the 30 titles recommended for such treatment with total ISI citations, 21 (70.0%) were from foreign commercial publishers, and 8 (26.7%) were those of U.S. commercial publishers. The same was the case with the 35 titles recommended for cancellation or nonpurchase with total UIUC use. Here, 24 (68.6%) were from foreign commercial publishers, and 10 (28.6%) were products of U.S. commercial publishers. Of these titles, 24 were common to all three cancellation or nonpurchase sets. One was a U.S. association serial, the Russian translation journal. It was the only U.S. association journal recommended for cancellation or nonsubscription, and it was consistently recommended for such treatment.
Although applicable in the management of serials collections, the very bases of the sociometric measures of ST value—particularly, total ISI citations —employed in the Evaluator's algorithms make them extremely dangerous for librarians from a political standpoint. This is because their utilization places the library at the very center of the promotion and tenure system. Symptomatic of this situation, as a result of his expertise in these measures, the principal author of this paper was asked by the LSU Graduate Council in the mid-1980s to write a report (Bensman 1985a) assessing the university's research-doctorate programs for possible termination. For this reason, it was decided to explore an accounting algorithm based upon cost-per-use for the Evaluator. The advantages of such an algorithm were perceived to be twofold. Not only would it enable the library to base cancellations on politically less dangerous measures, but it would also provide information directly related to decisions to transfer from ownership to access.
The same set of 120 chemistry titles was used to test the accounting algorithm, and serials were selected for cancellation purely on the basis of descending order of cost-per-use as calculated above until approximately the same cost reduction had been achieved as had been by the second algorithm with the sociometric measure of total UIUC Chemistry Library use—in this case, $77,498.15 (48.4%). The results spoke favorably of the cost-per-use method. Only with respect to number of titles canceled—44—did the accounting method fall behind the sociometric method, because many cheaper titles were brought up for cancellation, but in every other respect the accounting method proved to be more efficient. Thus, the percentage losses in value and the ratios of budgetary reduction to value lost were the following: total UIUC use—7.7%, a favorable ratio of 6.3 to 1 (48.4% to 7.7%); and estimated annual UIUC use—8.1%, a favorable ratio of 6.0 to 1. Moreover, the mean cost-per-use measures of the subset canceled by the accounting method—$594.67 by title, $246.42 by document—were higher than the respective measures in all three of the cancellation or nonpurchase subsets created with the sociometric methods.
As before, the commercial publishers bore the brunt of the cancellation or nonpurchase recommendations. Of the 44 titles recommended for such treatment, 28 (63.6%) were those of foreign commercial publishers, 14 (31.8%) belonged to U.S. commercial publishers, 1 was a foreign association title, and 1 —unsurprisingly, the Russian translation journal—was a U.S. association title. One-half of the titles in the subset recommended for cancellation or nonpurchase by the accounting method also belonged to the subset recommended for such treatment by the sociometric method based on total UIUC use. The accounting method recommended for cancellation or nonpurchase 18 (75.0%) of the 24 titles common to all three subsets created by the sociometric measures. However, before one waxes enthusiastic over the accounting method, it should be pointed out that the UIUC Chemistry Library use database was of extraordinarily high quality—even covering unbound issues—and it required no fewer than 23,412 counts to be made over a three-month period for its construction. One has to question whether many libraries—especially ones with broader subject scopes—have the organizational and labor potential to construct such databases.