SJR and SNIP : two new journal metrics in Elsevier ’ s Scopus

Journal citation measures were originally developed as tools in the study of the scientific-scholarly communication system1,2. But soon they found their way into journal management by publishers and editors, and into library collection management, and then into broader use in research management and the assessment of research performance. Typical questions addressed with journal citation measures in all these domains are listed in Table 1. Table 1 also presents important points that users of journal citation measures should take into account. Many authors have underlined the need to correct for differences in citation characteristics between subject fields, so that users can be sure that differences are due only to citation impact and Serials – 23(3), November 2010 Lisa Colledge, Félix de Moya-Anegón et al SJR and SNIP: two new journal metrics

not to the fact that journals sit in different subject fields that happen to have distinct citation characteristics 3 . Much has been published on the various approaches to journal ranking: Pinski and Narin 4 were the first to develop a journal citation measure (influence weight) that weights citations with the prestige of the citing journal. Recent studies explored usage-based journal impact metrics as additional indicators of journal performance 5 . Moed 6 underlined the importance to prolific authors of specialist journals often with relatively low journal impact factors. Finally, Garfield 7 , Seglen 8 and many others have underlined that journal impact factors should never be used as surrogates of measures of actual citation impact of individuals and research groups.
The journal impact factor, developed by Eugene Garfield as a tool to monitor the adequacy of coverage of the Science Citation Index, is probably the most widely used bibliometric indicator in the scientific, scholarly and publishing community. It is published in Thomson-Reuters' Journal Citation Reports (JCR) and will be labelled as 'JIF' in this paper. However, its extensive use for purposes for which it was not designed has raised a series of criticisms, all aiming to adapt the measure to the new user needs or to propose new types of indicators. (For reviews, the reader is referred to Glänzel and Moed 9,10 .) In January 2010, Scopus endorsed two such measures that had been developed by their partners and bibliometric experts SCImago Research Group, based in Spain and headed by Professor Felix de Moya, and the Centre for Science and Technology Studies (CWTS), based in Leiden, Netherlands, and headed until recently by Professor Anthony van Raan. The two metrics that were endorsed are SCImago Journal Rank (SJR) and Source Normalized Impact per Paper (SNIP). The aim of this paper is to provide an introduction to these two indicators. The following section gives a concise description of the methodologies underlying their calculation, and highlights their characteristics, comparing one with the other, and each of them with JIF. (For technical details about SJR and SNIP, the reader is referred to González-Pereira et al 11 and Moed 12 .)

Two newly endorsed journal metrics
Elsevier partners with several expert bibliometric groups around the world to ensure that their approach and developments are valuable and accurate. The following metrics have been developed for Elsevier by the SCImago Research Group and CWTS. The metrics are calculated by the bibliometric teams externally from Elsevier, using Scopus raw data supplied by Elsevier, and the values supplied are displayed in Scopus and other locations. Apart from customer demand, an important incentive for Elsevier to include journal metrics in Scopus is to stimulate more discussion on appropriate journal metrics and appropriate use amongst the people who use them in evaluations, and whose careers are affected by said evaluations.

SJR
One of the limitations of traditional citation analysis is that all citations are considered 'equal'. A citation from a widely-read, multidisciplinary journal counts as strongly as one from a more focused or local-interest source. SJR is a prestige metric inspired by Google's PageRank™, whereby a journal's subject field, quality and reputation have a direct effect upon the value of the citations it gives to other journals. The basic idea is that when a journal A is cited, say, 100 times by the most highly ranked journals in the field, it receives more prestige than a journal B that also receives 100 citations, but from less prestigious periodicals with a low visibility at the international research front. SJR makes a distinction between journal popularity and journal prestige. One could say that journals A and B have the same popularity, but A has a larger prestige than B. A and B would have the same JIF, but A would have a higher SJR than B.
Generalizing, JIF can be considered as a measure of popularity, as it sums up all citations a journal receives, regardless of the status of the citing journals, whereas SJR measures prestige. The idea of recursion, or iterative calculation, is essential.
Step by step, SJR weights citations in the current step according to the SJR of the citing journal in the previous step. Under certain conditions this process converges so that the SJR values do not change significantly any more with additional steps, and in the end a citation from a source with a relatively high SJR is worth more than a citation from a source with a relatively low SJR.
SJR also aims to limit excessive benefits derived from journal self-citation. When calculating SJR, the SCImago Research Group discount journal self-citations once they exceed one-third of the total citations received by a journal. In this way, the value of journal self-citations is still recognized, but SJR attempts to limit what are often seen as manipulative citation practices.

SNIP
SNIP corrects for differences in topicality between subject fields. It is a ratio of a journal's citation impact and the degree of topicality of its subject field. SNIP's numerator gives a journal's raw impact per paper (RIP), which is very similar to the JIF. Its denominator is the citation potential in a journal's subject field, a measure of the citation characteristics of the field the journal sits in, determined by how often and how rapidly authors cite other works, and how well their field is covered by the database (in this case, Scopus).
Citation potential can be conceived as a measure of the field's topicality. Fields with a high topicality tend to attract many authors who share an intellectual interest, and in this sense can be qualified as 'popular'. Developments in the field go quickly. Papers are written in a limited number of highly visible periodicals, and authors tend to cite -apart from the common intellectual base -the most recent papers of their colleagues. These popular fields will tend to have higher JIFs.
A journal's subject field is defined as the collection of articles citing the journal. Citation potential is calculated for this set of citing articles. This 'tailor-made' subject field delimitation takes into account a journal's current focus, and is independent of pre-defined and possibly out-ofdate field classifications. SNIP takes into account differences not only between, but also within journal subject categories. Table 2 shows a typical example illustrating the SNIP methodology. The mathematical and the molecular biological journal have very different raw impacts per paper (1.5 versus 13.0), but the citation potential in the former journal's subject field is much smaller than for the latter (0.4 versus 3.2). Compensating for this difference by dividing RIP by citation potential, the SNIP values of the two journals are almost identical.

SJR and SNIP
In fields in which practitioners publish mainly in journals that are covered by the citation database they use, there is a high immediacy of citation, and a clear structure in terms of core journals and more peripheral ones. SJR tends to indicate a journal's membership of this core more clearly than SNIP does. On the other hand, in fields that are heterogeneous in terms of subject coverage and citation practices, and fields in which journals are not the dominant publication channels, SNIP tends to evaluate a journal's citation impact more in context than SJR does. SJR, SNIP and JIF calculations give different ranges of values. In the analysis presented in Table 3, we have used RIP as a proxy for JIF, so that all journals in Scopus can be included in the information for each of these metrics, and not only for SJR and SNIP. For each indicator and main field, Table 3 gives the 10th, 25th (first or bottom quartile), 50th (median), 75th (third or top quartile) and 90th percentile of the distribution of indicator values among journals. Journals have been grouped into five main fields. Interesting observations from Table 3   For the reader interested in more technical details about the indicators, we refer to Table 4 and its legend, presenting an overview of the main differences between SJR, SNIP and the 'classical' JCR journal impact factor.  Table 4

. Comparison of the three journal metrics discussed in this paper
The choice for a publication window in SNIP and SJR of three years rather than two or five is based on the observation that in many fields citations have not yet peaked after two years, and in other fields citations have peaked too long before five years. Application of a window of three years is the optimal compromise to give fields in which impact matures slowly to reach its maximum while not penalising fields in which impact matures rapidly 15 . SJR and SNIP are not affected by 'free' citations to documents that are categorized as 'uncitable' and that, even when they are cited, are not counted in the measurement of a journal's number of published articles (e.g., Moed, 2005).
Eigenfactor 16 , not included in Table 4, is a journal impact measure to some extent similar to SJR, but showing the following main  Prestige is shared between differences in database coverage more citations are present in more journals, and is also across subject fields database. JIF does not correct redistributed so that more for differences in database prestige is represented in coverage between subject fields where the database fields coverage is more complete

Concluding remarks, and which metric should I use?
Tables 3 and 4 illustrate that the two indicators SJR and SNIP are to a certain extent complementary. Compared to the basic, JIF-like RIP, SJR tends to make the differences between journals larger, and enhances the position of the most prestigious journals, especially -though not exclusively -in life and health sciences. SNIP makes differences among journals smaller in all main fields, and is highest amongst journals in engineering & computer science.
In view of the heterogeneity of engineering & computer science and social sciences in terms of citation characteristics, citation potential and database coverage, for users interested in assessing journals in these fields, SNIP is probably more informative than SJR. On the other hand, if one considers that the heterogeneity in quality among journals due to Scopus' broad coverage heightens the need to account for the prestige of citing journals, SJR is more appropriate than SNIP.
If one considers that topicality is important in journal performance, in the sense that the best journals cover the most topical research themes, it is appropriate to focus on SJR rather than on SNIP. Moreover, if one considers that it is appropriate to weight citations on the basis of the status of the citing journal, SJR is a better indicator than SNIP. On the other hand, if one considers that journal impact and topicality or citation potential are two distinct concepts that should be measured and assessed separately, SNIP and its components, especially raw impact per paper and citation potential, are more informative.
The fact that Scopus introduced these two complementary measures reflects the notion that journal performance is a multi-dimensional concept, and that there is no single 'perfect' indicator of journal performance.