Despite the success of faculty score as a predictor of library use, there was one troubling feature that emerged from the analysis of its relationship to UnCover use, i.e., the low overlap between what was perceived as important by the faculty and what was being heavily borrowed through UnCover. Thus, among the 79 high UnCover use titles there were only 31 (39.2%) medium and high faculty score titles. This indicated that although faculty score was a highly accurate measure within the sphere perceived by the faculty, important processes were taking place in library use outside the perception of the faculty. Suspicions of this nature were confirmed when an inspection of the NOTIS circulation records of the titles on subscription at LSU Libraries but in the zero faculty score class revealed extremely high use of some of these titles. Cancellation of these titles on the basis of faculty score alone had the potential of extreme damage to the ST serials holdings of LSU Libraries. Moreover, Kleiner and Hamaker (1997, 367–68) studied UnCover use at LSU from November 1995 through May 1996 and revealed another flaw of faculty ratings: on an individual basis, the faculty tended to see importance where no importance existed. This flaw manifested itself in the extremely small overlap between the titles recommended for subscription by faculty members taken individually and those actually used one or more times through UnCover.
As a result of these findings, we decided to create a citation-based measure both to analyze the processes occurring in library use beyond the perception of the faculty and to serve as a corrective in decisions concerning cancellations and new subscriptions. Construction of total citation measures had proved to be a laborious process in the analysis of the data from the 1993 pilot project with the LSU Department of Chemistry, and further research was considered necessary on the time value of ST information before a more efficient method for constructing such a measure could be developed. Given the purpose of creating a corrective for faculty score, it was considered essential to develop the citation-based measure within the context of sets defined by subjects, and impact factor was the only citation measure in the JCRs that satisfied this condition.
The citation data collected for the study based upon the 1993 SRP pilot project with the LSU Department of Chemistry (Bensman 1996) was again analyzed to determine whether impact factor could be converted into a measure applicable to library use. As noted above, the main fault of impact factor is that it is corrected for size, whereas library use of a given title logically has to be heavily affected by its size. The first part of the analysis was to do a nonparametric Spearman rank-order correlation of total citations with impact factor, and predictably this correlation turned out to be low—0.38—due to the fact that size was inherent in the former and excluded from the latter. However, inspection of the plot of total citations against impact factor revealed 11 outliers arising from the higher average citation rates of review journals. When these outliers were excluded, the Spearman correlation of impact factor with total citations rose to 0.47, indicating a fairly good relationship of impact factor—or average citation rate—to total citation rate under these conditions.
Inspection of the plot revealed that if one divided the subject set ranked by impact factor at the median into two classes, the upper class would contain not only all the review journals—considered a plus from the library point of view—but also the vast majority of the serials highest in total citations. Such a method appeared ideal for the construction of citation-based, ordinally ranked classification variables to test against UnCover use with the chi-square test of independence and Kendall's tau-b. Accordingly, all UnCover titles were divided into three classification variables on the following basis: "not covered," those for which ISI provided no impact factor; "low," those below the median impact factor of their respective ISI subject groups; and "high," those above the median impact factor of their respective ISI subject groups. The 1994 SCI JCR—supplemented where necessary by the 1994 SSCI JCR—was utilized for this purpose both to bring this analysis into line with the above correlation analysis of faculty score with total citations and to test whether citation-based measures could be used to predict future library use because the UnCover data ran from mid-1994 to mid-1996. Because ISI sometimes lists a title in several subject groups, a policy was established to use the impact factor ranking of the largest of the subject groups. If there was any systematic error caused by this method, it was that titles that would have been in the high impact category in applied fields with low citation rates such as engineering would be found in the low impact factor category as a result of being transferred to pure research fields with higher citation rates.
Table 5 was constructed for the purpose of analyzing the relationship of impact factor to library use. This is another 3x3 contingency table using the impact factor classification variables developed above together with the same UnCover use classification variables that were employed to test faculty score. Once again, the null hypothesis of no association between the variables was rejected (chi-square=32.2, p<<0.0001). The significant Kendall's tau-b of 0.153 (p<<0.0002) again showed that this association was positive.
An examination of the observed frequencies against the expected frequencies in the different cells of the table shows that whereas high impact factor functioned very well as a predictor of library use, low impact factor performed less well. As a matter of fact, the performance of the low impact factor titles resembles that of the titles not covered by the JCRs. Thus, the observed frequencies of both the titles not covered in JCRs and the low impact factor titles were greater than the expected frequencies in the low UnCover use class, about as expected in the medium UnCover use class, and lower than the expected in the high UnCover use class. In contrast to this, the observed frequencies of the high impact factor titles were lower than the expected frequencies in the low UnCover use class, about as expected in the medium UnCover use class, but 1.7 times greater than expected in the high UnCover use class. These differences manifested themselves in the average use per title in the different impact factor classes and the total UnCover use for which these classes were responsible. Whereas the average uses per title of the serials not covered by the JCRs and in the low impact factor classes were respectively 2.2 and 2.9, the average use per title of the high impact factor serials was 5.7, or roughly double. Concerning total UnCover use, the 238 (33.4%) titles not covered in the JCRs accounted for 20.1% of UnCover use, and the 212 (29.8%) low impact factor titles were responsible for 23.5% of UnCover use. In both instances the percentage of total titles was higher than the percentage of total use. The opposite was the case with the high impact factor titles, which had 262 (36.8%) titles accounting for 56.4% of UnCover use. Moreover, the high impact factor titles covered the crucial high UnCover use class better than both the medium and high faculty score titles taken together. Of the 79 titles in the high UnCover use class, the high impact factor titles represented 50 (63.3%), whereas both the medium and high faculty score titles accounted for only 31(39.2%). From this analysis it is evident that the social stratification system of science and technology is operative in library use in a sphere beyond the perception of the faculty.
An extremely interesting picture emerges once the UnCover use of the 135 titles on current subscription at LSU Libraries is taken into consideration. Of these titles, 45 (33.3%) were high faculty score titles, and 109 (80.7%) were high impact factor titles. With their addition, the number of high faculty score titles rose from 29 (4.1%) of the sample titles to 74 or 8.7%, and the number of high impact factor titles rose from 262 or 36.8% of the sample titles to 371 (43.8%). Combined into one set without any overlap, the high faculty score and high impact factor titles represented 110 (81.5%) of the 135 titles on current subscription that affected UnCover use, and they accounted for 85.6% of the UnCover use caused by this set. As separate sets, the high faculty score titles—both on subscription and not on subscription—accounted for 18.9% of overall UnCover use, and the high impact factor titles—both on subscription and not on subscription—were responsible for 58.9% of overall UnCover use. This finding brings into sharp focus those of previous studies of interlibrary loan use, and its implication is clear: the influence of the science and technology social stratification system is so strong that it dominates not only internal library use but also interlibrary loan use. Good ST information is indeed a rare commodity.
The final test of faculty score was to investigate the internal use of the titles on subscription at LSU Libraries but found in the zero class of this measure. These were the titles in the working universe of the curriculum cores either not named by the faculty during the SRP or listed beyond the 45-title limit set in the questionnaire. They numbered 279 titles costing $101,997. For this analysis, use data were collected from the NOTIS circulation records on the volumes of these titles dating from 1988 through the end of 1996. These years were deliberately selected to make the backfile of these serials roughly equivalent to the backfile of the serials held by UnCover, whose holdings dated mainly from 1989 forward. However, whereas the UnCover use was over a 2-year period on roughly a 7.5-year backfile, NOTIS use was calculated to be the average over a 7.5-year period on a 7.5-year backfile, because use of unbound issues—and, it must be added, much in-house use—was not captured by the NOTIS circulation system. For comparative purposes, a formula was developed to standardize NOTIS use against UnCover use. This formula took into account not only the difference in time periods but also possible in-house use of the titles in the zero class of faculty score. Basically, it worked out to multiplying the recorded NOTIS use by 0.3.
Contrary to expectations, the use of the titles in the zero class of faculty score was found to be shockingly high, ranging up to 128 uses per title for a total of 1,967 uses. Only 56 (20.1%) of the 279 faculty score zero class titles had zero use. However, as usual, use was heavily concentrated on a few titles. Of the 279 titles, 80 (28.7%) accounted for 80.2% of the use, and 96 (34.4%) had zero or one use, amounting to 40 uses (2%) of the use. Predictably the variance of the NOTIS use of the faculty score zero class titles was found to be greater than the mean, rejecting the null hypothesis of randomness and indicating the presence of the NBD or a related contagious distribution.
Under these circumstances it was decided to investigate the processes operative in the NOTIS use of the titles in the zero class of faculty score by testing NOTIS use against impact factor. The method adopted for this was again the chi-square test of independence. NOTIS use was defined in two classification variables: "low" NOTIS use—the 199 (71.3%) of the faculty score zero class titles that accounted for 19.8% of NOTIS use; and "high" NOTIS use—the 80 (28.7%) of the zero class titles that accounted for 80.2% of this use. The entry level for high NOTIS use was 7 uses, which standardized into 2.1 UnCover uses, or roughly the dividing line between low and medium UnCover use. Such a division was considered appropriate, because neither NOTIS use nor the UnCover use of titles not on subscription at LSU Libraries concerned extremely high status titles such as Journal of the American Chemical Society, Nature, and Science.
Once NOTIS use variables had been defined, the 1995 SCI JCR was employed to construct two impact factor classification variables in the manner outlined above, except that both titles not covered by the JCR and low impact factor titles were lumped into one class. The result of this operation was 204 "low" impact factor titles and 75 "high" impact factor ones. Both the NOTIS use and impact factor variables could be ordinally ranked, and they were cast into a 2x2 contingency table with familiar results. The null hypothesis of no association between the NOTIS use and impact factor variables was resoundingly rejected by the large calculated chi-square of 24.3, a size with less than a one in 10,000 chance of occurring under conditions of the null hypothesis, and the Kendall tau-b was a positive 0.295 significant at below the 0.0002 level.
These results were verified by checking the observed frequencies against the expected frequencies in the table's cells. For the low impact factor titles, the observed frequency of 162 in the low NOTIS use class was above the expected frequency of 145.5, and their observed frequency of 42 in the high NOTIS use class was far below their expected frequency of 58.5. As for the high impact factor titles, their observed frequency of 37 in the low NOTIS use class was 30.8% below their expected frequency of 53.5, and their observed frequency of 38 in the high NOTIS use class was 76.7% above their expected frequency of 21.5. These findings once again confirmed that the social stratification system of science and technology influenced library use in a sphere beyond the perception of the faculty, making it necessary to apply correctives to faculty score in any decisions concerning serials cancellations and new subscriptions.
An inspection of the faculty score zero class titles underlying high NOTIS use revealed that two interrelated causal elements played a large role in the failure in faculty perception. One was the large and complicated bibliographic structure of the publications of the Institute of Electrical and Electronics Engineers (IEEE). The numerous and complex titles of the IEEE publications made them difficult both for the faculty to name as well as for the library staff to identify and score. This difficulty was compounded by the low response rate of the Department of Computer Science (4 of 15 faculty members, or 26.7%) to the SRP survey. Of the 80 titles accounting for 80.2% of NOTIS use, 8 titles were classed in the Computer Science curriculum core, and these 8 titles accounted for 312 (19.8%) of the 1,578 uses of the top 80 titles. All in all, there were 12 Computer Science titles with 317 total NOTIS uses in the faculty score zero class. Of these 12 serials, 6 were IEEE titles accounting for 147 (46.4%) of the 317 Computer Science NOTIS uses, and 4 were publications of the Association for Computing Machinery (ACM) accounting for 165 (52.1%) of the uses attributable to this curriculum core. Together the titles of these two U.S. associations were responsible for 98.5% of NOTIS uses in Computer Science. In addition to the Computer Science titles, there were two more IEEE titles that belonged to the Electrical Engineering curriculum core in the faculty score zero class, and these titles had a total of 11 NOTIS uses. Alone, the ACM and IEEE titles in Computer Science and Electrical Engineering numbered only 12 ( 4.3%) of the faculty score zero class titles, but they accounted for 16.4% of the NOTIS use of this class.