Quantifying the SRP survey data into a measure of ST value named "faculty score" involved two basic steps. First, the faculty were instructed to rank in order of importance up to 45 serials. These rank order lists were divided into quintiles with each quintile being assigned points from 5 to 1 in accordance with the descending order of the quintiles. If a faculty member listed more than 45 serials, the serials over 45 were included but assigned a score of zero. On the other hand, if a faculty member listed only a small number of journals—say 15 or less—all the journals were given the highest scores possible. Second, if faculty members designated a serial with an S (Subscription) as being necessary on campus, the serial was given an extra 5 points. If a serial was designated DD (Document Delivery), it was given no extra points. Thus the highest number of points a faculty member could give a serial was 10 for both being in the top quintile and necessary on campus. Dividing a serial's total faculty score by 10 made it possible to determine an equivalent number of professors who assigned that serial top priority. For example, the highest ranked serial was Science with a faculty score of 762, which was equivalent to having 76.2 professors give it top priority.

With the subject sets defined as curriculum cores and the measure of ST value quantified as faculty score, two databases were constructed. The first was called the "desired universe." It consisted of all those serials named by the faculty in the SRP survey and classed in one of the 33 curriculum cores derived from LC schedules Q, S, and T. Evaluator computer runs were made to determine the serials that were necessary to have LSU's serials holdings at 75% of the ST value perceived by the faculty in the desired universe of each curriculum core. The level of 75% of perceived ST value was chosen both because prior work with informetric distributions had indicated this as an optimal level and because anecdotal evidence suggested that such a level was politically palatable to the LSU faculty. Next, the desired universe of serials was checked to determine those that were on subscription at any of the libraries on the LSU campus. Furthermore, LSU Libraries' ST serials holdings also were investigated to discover those serials that were on current subscription in LC classes Q, S, and T, but not named by any faculty member in the SRP survey. Such serials were given a faculty score of zero and assigned to their proper curriculum cores on the basis of their LC call numbers.

From this information we constructed a second database called the "working universe," which consisted of all serials on current subscription at LSU libraries, including the zero class—i.e., either those not named at all by the faculty or those listed by them above the 45-title limit set in the SRP questionnaire—plus those serials not on subscription on the LSU campus but necessary to bring LSU's serials holdings up to 75% of the ST value perceived by the faculty in the desired universe of each curriculum core. The purpose of the exercise was to determine whether the working universe still contained enough resources after the massive cancellations of the 1980s and 1990s to bring LSU's serials holdings up to 75% of the ST value perceived by the faculty in the desired universe of each curriculum core.

Testing Faculty Score against Citations

However, before carrying out the exercise, the faculty scores were checked for validity by two different methods. The first method was to determine the correlation of faculty score with total ISI citation rate. This method was not just a test of how well the opinion of the LSU faculty corresponded to the research and social importance of serials in science and technology as a whole. It would also authenticate whether the LC classification system could be used to construct subject sets, as poor subject set definitions could lead to the inclusion of numerous outliers and negate any significant results.

The second method was to utilize the correlation technique to see whether faculty score was valid over time. Stability over time was considered essential because if it were nonexistent, the entire SRP could be considered an exercise in futility as any steps made to improve LSU Libraries' serials holdings in the present would be counterproductive in the future. High stability of the ST value distributions over time was expected due to the operation of the probability structure underlying the social stratification system of science and technology as well of the journal system based upon it.

For validating faculty score with the correlation technique, Chemistry, Agronomy, and Mechanical Engineering were again selected as the test curriculum cores. To obtain the serials for the tests, we used the faculty score lists generated by the Evaluator computer runs on the desired universe of serials. These lists ranked serials in descending order by faculty score, and a systematic sample for each test core was taken by selecting every third serial, starting with the highest ranked title. In this way we sought to capture samples representative of every stratification level of the test curriculum cores. The serials so chosen were compared to those covered in the 1985 and 1994 SCI JCRs. If a serial was not covered in both these JCRs, it was rejected, and another one was selected from the Evaluator lists among those near it in faculty score.

The 1985 JCR was picked because a decade was considered a reasonable amount of time for a test of stability, both in terms of the basic fairness of the test and of the effort required for a repetition of a project similar to the SRP. The 1994 JCR was selected because it was the latest one available at the time of the test. Total citation counts were derived for the selected serials for each year. The final sample sizes were the following: Chemistry–38; Agronomy–17; and Mechanical Engineering–13. Given the problems of set definition, errors inevitable in the collecting of such massive amounts of data, and the different subject emphases at LSU in comparison with science and technology as a whole, we decided that positive correlations of 0.50 and above would be considered a validation of the faculty score measures.

To prepare for the correlation tests, the nature of the probability distributions underlying the data was first investigated. As was expected, the variances of all the distributions were found to be significantly greater than the respective means, indicating the presence of the negative binomial. This called for the logarithmic transformation of the variables in order to use the parametric Pearson product-moment correlation. Therefore, the natural log or ln transformation of the data was performed.

With the proper mathematical transformation of the data implemented, plots and residuals were then examined for influential observations and influential outliers. Simply defined, an influential observation is one that plays a major role in determining the size of the correlation coefficient, and it may or may not be an outlier. To obtain the residuals, the correlations were treated as regressions with one independent variable. Where faculty score was involved, it was treated as the dependent variable, because it was assumed to have the most error, and in the correlations between total citations, the 1994 citations were made the dependent variable, because the logic of cumulative advantage dictated that they should be a function of the 1985 citations. As a general rule, the observations highest in faculty score and total citations were the influential ones, although this role was sometimes played by those at the lowest end of the distributions. This appears to indicate that both the nature and stability of the distributions appear to be anchored at the extremes of the distributions—a logical consequence of the double-edged Matthew Effect.

Two basic types of influential outliers were found. The first was the result of subject factors, and there were two of these, both of which manifested higher citations than warranted by their faculty scores. One of these was the agronomy title, American Potato Journal, which suggested that the topic involved was a specialty more important to the field as a whole than to the LSU faculty. The other subject outlier, Scripta Metallurgica et Materialia, affected Mechanical Engineering and represented a problem of set definition. By its original call number, this serial was classed in TN (Mining Engineering; Metallurgy), and it was published under the sponsorship of the American Society of Metals International and American Institute of Mining, Metallurgical, and Petroleum Engineers. This serial was one of the 12 TN titles artificially allocated among the curriculum cores on the basis of the departmental origins of the faculty selecting them, and its appearance in the Mechanical Engineering core thus appears to be a function of Garfield's law of concentration. It was decided to perform the correlations both with and without these subject outliers in order to check the effect of subject problems on the correlation coefficients.

The second type of influential outlier was the result of time factors, and involved the appearance of new journals. These affected the correlations of the 1985 total citations with the 1994 total citations, and the latter were much higher than the former due to the buildup of backfiles in the intervening period. There were again two of these. The first was a chemistry journal called Langmuir, which was started by the American Chemical Society in 1985. The second one was a mechanical engineering journal called International Journal of Robotics Research, which began publication by MIT Press in 1982. Given the connections of their publishers, the rapid rise of these journals in the citation ranks might perhaps be attributed to their being new extrusions of the ST elite. To handle these time outliers, it was decided to exclude them from the correlation of faculty score with 1985 citations, because it seemed illogical to compare their 1995 faculty score with their 1985 total citations under such conditions. However, with respect to the correlation of the 1985 and 1994 total citations, it was decided to retain them in the first correlations and then remove them in another correlation test to see how the appearance of new, elite journals affect the distributions of ST value over time.

In general, the correlation tests to validate faculty score and the stability of ST value distributions across time proved to be satisfactory. For Chemistry, the correlation of the 1995 faculty score with 1985 total citations was 0.51 (without the time outlier Langmuir) and 0.56 with 1994 total citations (with Langmuir). Concerning stability across time, with the inclusion of Langmuir, the correlation of 1985 and 1994 total citations in Chemistry was 0.85, which rose to a rock-solid 0.97 upon the exclusion of this time outlier. With respect to Agronomy, the correlation of 1995 faculty score with 1985 total citations was 0.61 and with 1994 total citations it was 0.58, including the subject outlier American Potato Journal. The exclusion of this subject outlier raised the correlation of the Agronomy faculty score to 0.71 with the 1985 total citations and to 0.66 with the 1994 total citations. As for stability across time, the correlation of the 1985 and 1994 total citations in Agronomy was a steady 0.85.

Mechanical Engineering serials manifested similar characteristics of fairly good correlations of faculty score with total citations and high stability across time. With the inclusion of the subject outlier Scripta Metallurgica et Materialia, the correlation of 1995 faculty score was 0.80 with 1985 total citations and 0.71 with 1994 total citations. These correlations rose respectively to 0.89 and 0.77 upon the exclusion of this subject outlier. A high stability across time manifested itself in Mechanical Engineering with a correlation of 0.96 between 1985 and 1994 total citations even with the inclusion of the time outlier International Journal of Robotics Research. The exclusion of this time outlier raised this correlation a small bit to 0.98. What is most interesting is that the correlations of 1995 faculty scores were in the same range with 1985 total citations as with 1994 total citations, again demonstrating the stability of the ST value distributions across time. All the above correlations were significant at the 0.05 level.