Exact fits in bibliometrics : Some tools an results

Abstract

The present paper takes an almost diametrically opposed approach to that of recent work in bibliometrics, such as Sichel, which has been concerned with increasing formalization. The reported series of experiments start with versions of the simplest bibliometric models utilizing few, and explicit parameters, and then, attempts to determine how much structure in very large data sets can be readily and simply explained. A system for the creation of empirical mathematical models was developed, consisting of I) software for microcomputers and 2) a basic discipline of analysis. Good fits to data were obtained and replicated for large ranked distributions, for Lotka Law distributions, and for literature growth and aging phenomena. Several interesting results were identified and displayed, including: I) In very large ranked distributions (ranks > 3,000), the rate. of change between adjacent ranks varied across the distribution as a function of rank, changing average ca. 0.001 every 180 ranks. Other studies typically use data sets t h a t a r e too small f w this change to be detected. 2) The most used sources, in these distributions, present a special case, one that' might not be fit by any of the candidate functions, suggesting a new, non-arbitrary definition of "core" literatures. 3) It was easy, using this approach to develop fits to Lotka distributions t h a t a r e demonstrably better than Pao's, and thereby, to detect a flaw in her opproach. 4) The approach allowed "what if" analyses of library collection development f w the planning of acquisitions and the identification of space and storage requirements.