The second method of validating faculty score was to check its relationship to UnCover use. An analysis of faculty score as an indicator of library use was considered essential, because previous researchers had given mixed reviews to expert ratings as predictors of library use. For example, Wenger and Childress (1977) found at the National Oceanic and Atmospheric Administration Library that a journal fell into the low-use category 13.7% of the time when recommended by one or more scientists, but only 5.5% of the time when recommended by two or more scientists. However, Bustion and Treadwell (1990) concluded from study done at Texas A&M that a high ranking of a serial by the faculty did not prove to be a predictor of high use. In their opinion, there appeared to be a very weak relationship between the value of a periodical perceived by the faculty and its subsequent use.

Before undertaking the analysis of faculty score as an indicator of UnCover use, it is necessary to clarify the true nature of the interlibrary use of serials. Interlibrary loan use of serials is not the random use of rare and unimportant titles. On the contrary, it manifests the same characteristics of serials use within a library and is dominated by the same titles. This is evident in the report by Pilling (1986) on a study of 1983 serials demand at five major document supply centers in Europe and the U.S.—British Library Document Supply Centre (BLDSC), Centre de Documentation Scientifique et Technique (CDST), Chemical Abstracts Service (CAS), National Library of Medicine (NLM), and Online Computer Library Center (OCLC)—in which it was found that 70% of requests for all the organizations were met by between 15% and 32% of the titles. Moreover, despite the widely differing functions of the five document supply centers, there was a fair amount of overlap among their highly requested titles, and it was possible to construct a model "composite" document supply center where a mere 514 titles would meet 20% of the composite demand.

The first U.S. analysis of the interlibrary loan use of ST journals was done by Kurth (1962) with respect to the 77,698 requests made to the National Library of Medicine (NLM) by 1,780 domestic and foreign libraries during 1959. Although the NLM held around 37,000 serials titles, it required only 4,347 titles to fill 100% of the requests, meaning that the 1959 NLM zero class was around 88%. Of the 4,347 titles borrowed in 1959, 161 titles (3.7%) accounted for 30,203 (38.9%) of the loans. Among the top 15 titles borrowed from the NLM in 1959 were such present-day stalwarts as Lancet, British Medical Journal, Journal of the American Medical Association (JAMA), Journal of Biological Chemistry, New England Journal of Medicine, Science, and Nature.

A similar pattern of interlibrary loan use manifested itself in a study conducted during the 1960s at the Chemical Abstracts Service (CAS) of the American Chemical Society (ACS) to determine the availability in the U.S. of primary ST documents in serials and conference proceedings. As reported by Wood (1969), this study naturally focused on chemistry and chemical engineering literature, but its results were found applicable to other ST disciplines. Unlike the NLM study, the CAS study was not an analysis of requests to a single document supply center, but it was based on a sample of 70,686 interlibrary loan requests provided by 19 resource libraries distributed evenly throughout the U.S. These requests originated from persons in some 3,363 U.S. organizations. The bulk of them were made in 1967, although some related to the latter part of 1966 and first part of 1968. The requests were for documents in 12,282 serial and conference proceeding titles, and, once again, a small percentage of the titles provided a large part of the documents. Thus, 195 (1.6%) of the titles accounted for 17,777 (25.1%) of the requests; 850 (6.9%) of the titles accounted for 35,671 (50.5%) of the requests; and 2,662 (21.7%) of the titles accounted for 53,102 (75.1%) of the requests.

Included in Wood's report was a ranking of the 850 titles accounting for 50.5% of the requests. It was a multidisciplinary list, and the top chemistry title ranked third in number of interlibrary loan requests. This title was the Journal of the American Chemical Society—the title with the greatest faculty score in desired universe of the Chemistry curriculum core resulting from the LSU SRP survey in 1995. Of the top 15 titles accounting for some 50% of the faculty score in this core, 10 were in existence in 1967. All 10 were among the 850 titles listed by the ACS study as highest in interlibrary loan requests. These 10 titles represented 0.08% of the titles in the ACS study but accounted for 1,241 (1.8%) of the interlibrary loan requests, a favorable ratio of about 22.5 to 1. Four of these 10 titles were U.S. association journals—3 of the American Chemical Society and 1 of the American Institute of Physics. Of the 5 titles not in existence in 1967, 3 were subsequently established by the American Chemical Society.

The findings of these U. S. analyses of interlibrary loan use were replicated in a series of studies conducted at the British Library Lending Division (BLLD), the former name of the present British Library Document Supply Centre (BLDSC). Although the BLLD was a central interlibrary loan library, the use of its holdings appeared to be typical of that of any large general academic library, and a close relationship was found to exist between use at the BLLD and at Newcastle University Library. The first BLLD study was done in 1975 and was based on a sample of 61,333 serials requests constructed by selecting every sixth request during the first three months of that year (BLLD/SINFDOK 1975; Bower 1976; Line and Wood 1975). Of these requests, 59,617 were for 14,718 serials titles held by the BLLD, and a further 1,716 requests were for titles not held there. As usual, demand concentrated on a very few titles, leaving a large zero class. In the case of requests for titles held by the BLLD, 50% were in just over 1,300 titles in spite of the fact that BLLD was then currently receiving 45,000 titles and held more than 100,000 titles altogether. Only 34% of the requested titles accounted for 80% of the demand. This 34% comprised a mere 10% of titles then currently received by the BLLD and represented only 5% of all titles held.

An extremely interesting phenomenon was that the degree of use concentration differed in science and technology, the social sciences, and the humanities, perhaps reflecting the differing structures of these areas in terms of Kuhnian paradigms. Whereas science and technology requests were concentrated on a relatively few titles (50% of the requests were in only 8% of the requested titles in this field), social science requests were rather more widely spread (50% of them being in 12% of the requested social science titles), and the humanities requests were even more dispersed over the collection (50% being in 21% of the requested humanities titles).

However, what puzzled the researchers the most was the nature of the highly requested titles. Nearly all were in science, and "pure" science at that; many had large circulations; all appeared to be high-status journals; and most were widely held by British libraries. Line and Wood (1975) constructed from the survey data a list of 81 titles that by extrapolation would have had 300 or more requests annually over the preceding three years, and a glance at the top 15 titles on this list reveals what looks to be a roundup of the usual suspects. These 15 included not only the four medical titles most highly requested from the NLM in 1959— British Medical Journal, JAMA, Lancet, and New England Journal of Medicine—but also such titles as Analytical Biochemistry, Analytical Chemistry, Biochimica et Biophysica Acta, Journal of Biological Chemistry, Journal of Chromatography, Journal of the American Chemical Society, Nature, Proceedings of the National Academy of Science, Science, and Scientific American that were highly ranked by the LSU faculty in the 1995 SRP survey. The most requested social science title at the BLLD in the 1975 sample was the American Sociological Review, and the most requested humanities title was Past and Present (Bower 1976, 33–34).

A second study of interlibrary loan use was conducted by the BLLD in 1980. As described by Clarke (1981), in contrast to the 1975 survey, the second study was not based on a sample of every sixth request gathered over a three-month period but on all valid serial requests collected over 10 consecutive working days in May 1980, which totaled 66,430 requests for 18,975 titles. Results from the 1980 survey revealed the same characteristics of interlibrary loan use as had the 1975 study, and two lists were produced that ranked titles by order of the requests for them in their respective years. A comparison of the two lists revealed an apparent instability in library use. Thus, there was only a 60% overlap among the top 100 titles requested in 1975 and 1980, and this overlap gradually decreased as one went down the ranks until only 2,591 titles (52%) were common to the top 5,000 titles on both lists. Clarke (1981, 111) summed up the conclusion derived at the BLLD from this overlap analysis thusly: "This inconsistency of rank lists sheds doubt on the continuing value of core lists of serials, which might decrease substantially in validity over a relatively short period."

The melancholy conclusion reached at the BLLD as to the validity of core lists of serials over time drew a hilarious response from Urquhart (1982), who in the role of Sherlock Holmes, set out to solve the statistical crimes committed at this institution. In going over the BLLD overlap analysis, Urquhart found a number of major faults in method and reasoning: (1) changes in title status such as births, deaths, marriages, and divorces were not taken into account; (2) the 1975 and 1980 sample sizes were different; (3) the sampling periods were different, raising the possibility that variation in use levels might have been greater because of peaking of demand over short periods; and (4) natural variations can be expected from one survey to the next without any real change in behavior, and these variations can be surprisingly pronounced for individual titles. Even without taking into account possible peaking effects in short time periods, Urquhart calculated that the expected overlap of the upper 5,000 titles in 1975 and 1980 use could only have been 64%, and he pointed out that this compared favorably to the 52% figure reached by BLLD. Urquhart therefore rejected the conclusions reached at the BLLD regarding the instability of library use.

In 1983, a third study of the interlibrary loan requests was conducted at the BLLD. As reported by Merry and Palmer (1984), the methods used in 1983 were the same as those in the 1980 study, and the sample consisted of all serials requests for ten working days in May. This time the number of requests amounted to 66,720, of which 61,946 were for 18,465 titles held by the BLLD. The results were basically the same, and a relatively small number of serials—2,158—accounted for 50% of the demand. In one interesting measure, it was determined that the concentration of use had increased over the years and that the percentage of titles (current or ceased) satisfying 100% of demand shrank from 15% of all titles held by the BLLD in 1975 to 14% in 1980 to 11% in 1983.

Another overlap analysis—this time between lists ranked by interlibrary loan requests in 1980 and 1983—revealed that about 60% of the titles were common to both lists, regardless of whether the top 100 or the top 5,000 were compared. Despite Urquhart's riposte, the apparent instability of interlibrary loan use indicated by these overlap figures continued to concern the BLLD staff, and a similar overlap analysis was conducted with respect to rank lists of journals constructed from citation data from different years of the SCI and SSCI JCRs. The JCR overlap percentages were much higher than the BLLD overlap percentages, and the BLLD director, Line (1984), speculated on the reasons for this, calling for more research into this matter.

Here two points should be made. First, the studies at the BLLD were done on a global basis, without any breakdown into subject sets, and a much higher pattern of stability might have been found within the subjects sets once possible surges among different subject groups had been controlled for. Second, even if library use of ST literature were more unstable than that of its citation patterns, library use—as the subject interests of the patron population shift—might only move within the overall stable framework of ST literature as this framework is defined by the social stratification system of science and technology being measured by citations.

This review of earlier studies of interlibrary loan use—with its similarity to internal library use, its concentration on well-known titles, its large zero classes, its apparent stability over time—indicates that there may be a definite bottom to ST serial literature and that this bottom may not be very deep. This bottom may have been found at both the University of Illinois at Urbana–Champaign (UIUC) Chemistry Library and Montana State University (MSU) Libraries. Concerning UIUC, Chrzastowski (1991) and Chrzastowski and Olesko (1997) describe three usage studies undertaken at the Chemistry Library there in 1988, 1993, and 1996. During this period, the UIUC Chemistry Library canceled over 180 chemistry journals or approximately 25% of its serials holdings. Despite the massive cancellations, the serials holdings of the UIUC Chemistry Library remained relatively unscathed. This was established by two studies—one of document deliveries and the other of interlibrary loan use.

Chrzastowski and Anthes (1995) conducted the first study. For six and one-half months from October 15, 1993, to April 30, 1994, the UIUC Chemistry Library experimented with supplementing its serials collection with document deliveries from the Chemical Abstracts Service (CAS). Of the 234 requests for documents, 176 were for articles from 136 journals. The majority of the requested titles—111 (81.6%)—had never been owned by UIUC, and these titles accounted for 145 (82.4%) of the requested articles. Only 31 (17.6%) of the requests were for articles from the 25 titles (18.4%) that had been previously canceled by the university.

The fundamental soundness of the serials holdings of the UIUC Chemistry Library was corroborated by Chrzastowski and Olesko (1997) with a title-by-title analysis of 1996 interlibrary loan requests made from this library. Of the 94 journals requested, 72 (76.6%) were titles never owned by the UIUC Chemistry Library, and from 64 of these serials only a single article each was requested. Seven titles were requested 2 times, and 1 was requested 3 times. Of the remaining 22 journals held by the UIUC Chemistry Library and requested on interlibrary loan, only 8 (8.5% of the total sample of 94) were titles previously canceled; 13 (13.8%) were for issues out at the bindery or otherwise unavailable; and 1 (1.1%) title was a recent subscription for which the library lacked the early volumes.

The lack of damage to the UIUC Chemistry Library serials holdings from the cancellations appears to be largely the result of the stability of the skewed distributions of journal use over time. For example, the top 10 journals accounted for 32.9% of the use in 1988 and 38.9% of the use in 1996, a rise of 18.2%. Eight of the top 10 were the same in both 1988 and 1996. The 2 titles dropping out of the top 10 fell only two places to numbers 11 and 12, and these 2 were replaced by titles previously in the top 15 (Chrzastowski and Olesko 1997).

For their part, MSU Libraries undertook a project in the cooperative collection of science serials with four other universities in the Pacific Northwest. This project was described by Price and Carey (1993), who analyzed some of its results. Implementing this project, MSU Libraries purchased 86 science serials with a pledge to make them readily available to the other universities. Both local and interlibrary loan use was monitored on 84 of these titles. Of the 84 titles, 30 (costing a total of $10,350) had no local use, and—to the evident surprise of the researchers —no interlibrary loan use, either. A number of calculations indicated that it would have been far more cost effective to have utilized document delivery through UnCover, and serious doubts were raised about the wisdom of the cooperative collection of science serials.

Having clarified the general nature of interlibrary loan use, it is now possible to examine with greater understanding the findings on the relationship of LSU faculty score to UnCover use. The sample for this analysis included all documents delivered to LSU Libraries by UnCover between July 1, 1994 and June 30, 1996 from serials classed in LC subject groups Q, S, and T. From these titles were weeded all those that were classed in parts of LC class schedules Q, S, and T that did not pertain to the curriculum cores under consideration. The resulting sample comprised 847 serials accounting for 2,909 document deliveries.

Despite titles on current subscription at LSU Libraries being blocked from the UnCover system, there was a considerable amount of leakage in the system. Of the 847 titles from which UnCover delivered documents in the two-year period, 135 (15.9%) were on current subscription at LSU Libraries, and these 135 titles accounted for 250 (8.6%) of the document deliveries. These titles were also weeded, leaving a final sample of 2,659 documents delivered by UnCover from 712 titles classed in LC subject schedules Q, S, and T during the study period.

To save labor, it was decided to test faculty score against UnCover use on a global basis, i.e., without breaking the UnCover titles into subject sets defined by the curriculum cores. This decision dictated statistical techniques not only resistant to outliers but also unaffected by the highly skewed distributions basic to any set of library data. Two such tests—both nonparametric—were chosen. The first was the chi-square test of independence (sometimes called the chi-square test of association or homogeneity), which operates on the nominal level (Snedecor and Cochran 1989, 76–79, 124–29, and 196–212; Hatcher and Stepanski 1994, 155–70; Stokes, Davis, and Koch 1995). For this test, "classification" variables are constructed by grouping the data into broad classes or groups. Then within these classes "expected" frequencies are calculated and compared to actual or "observed" frequencies. The null hypothesis is that there is no association between the variables, and this is tested by calculating and summarizing the differences between the "expected" and "observed" frequencies in a statistic called chi-square. If the chi-square is small, the null hypothesis of no association is accepted; if the chi-square is large, the null hypothesis of no association is rejected, and there is a high probability that the variables are related. The second statistical method chosen to analyze the relationship of faculty score to UnCover use was Kendall's tau-b, which operates at the ordinal level (Gibbons 1993; Schlotzhauer and Littell 1987, 371–76). Kendall's tau-b is similar to the Spearman rank correlation coefficient in that its values range from -1 to 1, with -1 indicating a strong negative association, zero showing no association, and 1 demonstrating a strong positive association between the variables.

For purposes of the test, UnCover use was grouped into three ordinally ranked classes. The first class, "low," consisted of the 310 titles (43.5%) that had been used to supply one document each for a total of 310 (11.7% ) of the 2,659 documents requested. The second class, "medium," consisted of those 323 titles (45.4%) that had been used to supply 2 to 6 documents, summing up to 977 (36.7%) of the documents requested. The third class, "high," contained the 79 titles (11.1%) that had been used to supply 7 to 198 documents each for a total of 1,372 (51.6%) of the documents requested. Given such a structure, it was not surprising that the variance of the UnCover use was found to be significantly greater than the mean, rejecting the null hypothesis of randomness and indicating the effect of the stochastic processes of qualitative inhomogeneity and contagion.

Faculty score was also divided into three ordinally ranked classes. These classes were based upon the rank of their constituent titles after the latter had been arranged in descending order by faculty score within each of their respective curriculum cores in the desired universe. The first class, "zero or low," comprised those titles that either had not been named by the faculty in the SRP survey, had been listed beyond the 45-title limit set in the survey questionnaire, or were in the bottom 25% of the aggregate faculty score of their respective curriculum cores. The second class, "medium," consisted of those titles that had an aggregate faculty score that placed the titles in the 51% to 75% range of the aggregate faculty score of their respective curriculum cores. The third class, "high," comprised those titles that accounted for the top 50% of the aggregate faculty score of their respective curriculum cores.

Here something important must be pointed out. Due to the high positive skew and exponential structure of informetric distributions, the interval distances between elements ordinally ranked in descending order by any informetric measure rapidly decreases as one goes down the ranks until there is little or no difference in absolute terms between the elements falling just above or just below the divide separating the top 75% from the bottom 25% of the aggregated informetric measure. Couched in economic terms and applied to ST serials, this dictates that the marginal utility of ST serials—or the utility added by each additional serial—diminishes with brutal rapidity. For example, on the average it took 9.9 titles (15.6%) to account for the top 50% of the aggregate faculty score in each of the 33 curriculum cores, but it took an average of another 11.9 titles (18%) to increase the aggregate faculty score another 25% to 75%.

From this perspective, it was possible to make a preliminary assessment that LSU Libraries' ST serials holdings were not heavily but only moderately damaged despite a full decade of adding no new subscriptions and undergoing massive cancellations. Of the 326 titles perceived by the faculty in the SRP survey as necessary for LSU Libraries' serials holdings to be at 50% of the ST value of the desired universe of all 33 curriculum cores, 53 or 16.3% were not on current subscription.

The results of the analysis of the relationship of faculty score to UnCover use are shown in table 4, which is a 3x3 contingency table constructed from the above classification variables. This table shows that faculty score is definitely associated with UnCover use. First, the calculated chi-square of 48.7 was highly significant (p<<0.0001), and, second, the significant Kendall's tau-b of 0.181 demonstrates that this association was positive. An inspection of the expected and observed frequencies in the various cells of the table corroborates this conclusion. Thus, the observed frequency of zero or low faculty score titles is higher than expected in the low UnCover use class, approximately the same as expected in the medium UnCover use class, and lower than expected in the high UnCover use class. In contrast, the observed frequency of medium faculty score titles was lower than expected in the low UnCover use class, about as expected in the medium UnCover use class, and more than twice as expected in the high UnCover use class.

High faculty score titles manifested the same behavior as medium faculty score titles, only more so. The observed frequency of high faculty score titles was less than half than expected in the low UnCover use class, about as expected in the medium UnCover use class, and about 3.8 times higher than expected in the high UnCover use class. Twelve (41.4% ) of the 29 high faculty score titles were in the high UnCover use class. The tendency of the medium and high faculty score titles to cluster near the top of the UnCover use distribution can be seen in the following average uses per title of the different faculty score classes: about three uses per title in zero or low faculty score class, 5.3 in the medium faculty score class, and 15.3 in the high faculty score class.

The average use per title in the medium and high faculty score classes was artificially depressed because many of them (36.7% in the medium, 55.2% in the high) had backfiles at LSU Libraries even though they were not on current subscription. When those with backfiles were excluded, the average use per title rose to 6.9 in the medium faculty score class and to 27.2 in the high. Although the total number of medium and high faculty score titles was 108 (15.2%) of the 712 UnCover titles, they accounted for 865 (32.5%) of the documents supplied in the two-year period. The 29 high faculty score titles alone accounted for 4.1% of the titles but 16.7% of the use—more than the 79 medium faculty score titles, which accounted for 11.1% of titles and 15.8% of the use.