Page - (000251) - in Knowledge and Networks

247 The first measure is a summary statistic for the central tendency of the combinations of journals that a paper cites. The larger the median z-score for a paper, the more common the main mass of journal combinations in that paper compared to chance. The second measure is a summary statistic for the left tail of combinations of jour- nals that a paper cites—journal pairings that are relatively unusual, compared to chance, among the set of journal pairings in that paper’s reference list. Illustrative Example of Methodology and Further Detail To illustrate these procedures, consider the following example, based on a single paper in the field of geography. 1. Step 1. Take the references in a bibliography in a given paper. Consider the paper, “The Tropical Cyclone Hazard Over the South China Sea 1970–1989: Annual Spatial and Temporal Characteristics,” which was published in Applied Geography in 1995. This paper has 22 references, of which 10 are known refer- ences (Fig. 12.2). 2. Step 2. Consider all pairwise combinations of the papers referenced in the bibli- ography of that paper. As can be seen in Fig. 12.2, pairwise paper combinations include, for example, (i) Deser et al. 1992 with Black 1990, (ii) Deser et al. 1992 with Thompson 1987, and (iii) Thompson 1987 with Black 1990. With 10 known references, we have 45 (i.e., 11 choose 2) pairwise paper combinations. 3. Step 3. Map the observed paper pairs into observed journal pairs. The 45 paper pairs are mapped into 45 journal pairs, where some journal pairs in this list appear multiple times. For example, Nature and Monthly Weather Review are paired twice. 4. Step 4. Repeat steps (1–3) for every paper in the WOS. The above steps, shown in a single article, are now repeated for every paper in the WOS. References to materials outside the WOS (for example, books) are not included. 5. Step 5. Count the frequency of each observed journal pairing for a given publica- tion year, using the referenced works of every paper published that year in the WOS. Information from the sample paper above would be counted as part of the year 1995. Hence, we allow journal pair frequencies varying over time. Having completed steps (1–5) for the observed papers in the WOS, we repeated them for each synthetic instance of the WOS, as created by the null model. Comparing the observed frequency of journal pairs under the real WOS with the frequency distribution that appears across instances of the null model, we computed a z-score for each journal pair. Continuing our illustrative example, the observed frequency, expected frequency, and z-score for several journal pairings that appear in the paper, “The Tropical Cyclone Hazard Over the South China Sea 1970–1989: Annual Spatial and Temporal Characteristics,” are presented in Table 12.1. As Table 12.1 demonstrates (for a subsample of journal pairs), each published paper has a distribution of journal pairs, some of which are highly conventional (such as 12 How Atypical Combinations of Scientific Ideas Are Related to Impact:…