Page 179 - Becoming Metric Wise

171
Journal Citation Analysis

millions of times. The relative number of times this “random journal
surfer” visits a journal is this journal’s eigenfactor. Clearly one needs an
extra rule in case a reference list is empty or does not contain a single
suitable item. One must also take precautions against infinite loops.
The AIS is a measure for the influence of a journal, normalized in
such a way that the average journal in the database has an AIS-value of 1
(details are provided in Subsection 6.9.2). If a top journal has an AIS of
12 this means that the average article in this journal has 12 times more
influence than the average article in the database. As all citation measures
also this measure is field dependent. In 2012 the journal with the highest
AIS in the JCR was Reviews of Modern Physics with a value of 32.565.
Among the 85 journals in the category Information Science and Library
Science only 8 had an AIS value larger than 1.

6.9.2 Calculating the Eigenfactor Score and the Article
Influence Score
In a previous section we explained the Pinski-Narin approach and pointed
out that in practice some adaptations are needed. Here we go into the practi-
cal calculation of the Eigenfactor Score and the AIS as described in http://
www.eigenfactor.org and Franceschet (2010a). For the mathematical princi-
ple we refer the reader to Langville and Meyer (2006). Let C 5 (C ij )be a
journal citation matrix for the year Y. The aim is to find the Eigenfactor
Score for each journal described by this citation matrix. The value C ij in the
matrix cell of the i-th row and j-th column denotes the number of references
given in journal i in the year Y to articles published in journal j during the
previous 5 years. This matrix is clearly not symmetric: in general C ij 6¼C ji
(although equality may happen occasionally for some i and j). Moreover, C ii
is set equal to zero, i.e., journal self-citations are omitted. Next we introduce
the article vector a. The components of a, a i , are equal to the number of arti-
cles published in journal i during the previous 5 years (the citation window),
divided by the total number of articles published by all journals represented
in C, during the same 5-year period. Hence, the component a i represents
the relative publication contribution of journal i in the journal network
under study. A dangling node (journal) i is a node that does not cite a single
other journal of the network. In a next step matrix C is transformed into a
normalized matrix H 5 (h ij ). In this matrix rows (journals) not correspond-
ing to dangling journals are divided by the row totals (so that the new row