Source-linked AI summary
Some modifications to the SNIP journal impact indicator
Ludo Waltman, Nees Jan van Eck, Thed N. van Leeuwen, Martijn S. Visser
TL;DR
The paper addresses counterintuitive properties of the original SNIP indicator, including decreases after additional citations and lower values after journal mergers. It introduces a revised normalization mechanism and finds that the revised and original indicators differ relatively little overall, while showing systematic ranking differences.
Problem
The original SNIP indicator has counterintuitive theoretical behaviors, and source normalization must also handle citing journals with special referencing behavior.
Method
The paper revises SNIP by changing its normalization mechanism, including harmonic DCP averaging, publication-proportion weighting, and revised citing-journal selection.
Results
The revised SNIP rules out the identified counterintuitive behaviors, while empirical differences from the original are relatively small and computer science and engineering journals tend to rank lower.
Takeaways & Limitations
The revised indicator preserves source normalization while addressing the original indicator’s cited theoretical problems and revealing systematic ranking differences.
Takeaways & Limitations
Because SNIP averages citation scores, highly cited outlier publications can substantially increase a journal’s value.
Abstract
from arXiv · showhide
The SNIP (source normalized impact per paper) indicator is an indicator of the citation impact of scientific journals. The indicator, introduced by Henk Moed in 2010, is included in Elsevier's Scopus database. The SNIP indicator uses a source normalized approach to correct for differences in citation practices between scientific fields. The strength of this approach is that it does not require a field classification system in which the boundaries of fields are explicitly defined. In this paper, a number of modifications that will be made to the SNIP indicator are explained, and the advantages of the resulting revised SNIP indicator are pointed out. It is argued that the original SNIP indicator has some counterintuitive properties, and it is shown mathematically that the revised SNIP indicator does not have these properties. Empirically, the differences between the original SNIP indicator and the revised one turn out to be relatively small, although some systematic differences can be observed. Relations with other source normalized indicators proposed in the literature are discussed as well.
1. Introduction
SNIP is a source-normalized journal citation-impact indicator designed to correct for field differences without explicitly defining field boundaries. The paper explains planned revisions, their advantages, and comparisons with other source-normalized indicators.
- 1. Introduction: SNIP measures scientific-journal citation impact using source normalization introduced by Henk Moed in 2010.The indicator is included in Elsevier’s Scopus database.
- 1. Introduction: Source normalization corrects for differences in citation practices between fields without requiring an explicit field-classification system.Normalization uses characteristics of the sources from which citations originate.
- 1. Introduction: The paper explains modifications to SNIP, discusses advantages of the revised indicator, and empirically compares it with the original.The paper also discusses other source-normalized indicators proposed in the literature.
- 1. Introduction: The paper proceeds from the original indicator and its properties to the revised indicator, citing-journal selection, empirical analysis, and conclusions.Technical results are presented in two appendices.
2. Original SNIP indicator
The original SNIP divides raw impact per paper by relative database citation potential to normalize citation impact across fields. Although this mechanism can work under appropriate assumptions, it has counterintuitive behaviors involving additional citations and journal mergers.
- 2. Original SNIP indicator: The original SNIP is defined as a journal’s raw impact per paper divided by its relative database citation potential.RIP is the average number of citations received by publications from the preceding three years, while RDCP normalizes database citation potential.
- 2. Original SNIP indicator: The database citation potential averages active references in a journal’s subject field, which consists of publications citing the journal’s recent publications.Active references point to publications from the preceding three years in journals covered by the database.
- 2. Original SNIP indicator: Under suitable assumptions, dividing RIP by RDCP can cancel field differences because citation-rich fields have both higher RIP and higher citation potential.The example contrasts fields averaging 12 versus 3 active references and corresponding average RIP values of 4 versus 1.
- 2. Original SNIP indicator: An additional citation can decrease the original SNIP when it comes from a publication with many active references.In the example, the journal’s SNIP falls from 4 to approximately 3.12 after receiving one citation with 100 active references.
- 2. Original SNIP indicator: Merging two journals can produce a lower original SNIP than either original journal because the merged DCP is not an average of their DCP values.The example gives the merged journal a SNIP of 5.40, below both original journals.
- 2. Original SNIP indicator: The revised SNIP is introduced to remove these two unsatisfactory properties of the original indicator.The paper states that the revised indicator does not permit these types of behavior.
3. Revised SNIP indicator
The revised SNIP changes the normalization mechanism while retaining the original RIP numerator. Its main modifications use harmonic DCP averaging, account for the proportion of citing-journal publications with active references, and produce more intuitive merger behavior.
- 3. Revised SNIP indicator: The revised indicator calculates DCP values as harmonic rather than arithmetic averages.The paper motivates this alternative average as more appropriate in some situations.
- 3. Revised SNIP indicator: The revised DCP calculation incorporates both active-reference counts and the proportion of publications with at least one active reference in citing journals.This proportion is denoted by p_i and is calculated from publications in the same journal and year as each selected publication.
- 3. Revised SNIP indicator: The revised SNIP changes only the normalization denominator, leaving the RIP numerator unchanged.RIP is calculated in the same way as for the original indicator.
- 3. Revised SNIP indicator: The revised indicator allows multiple occurrences of the same publication in a journal’s subject field and uses a three-year citation window.The three-year window matches the citation window used for RIP.
- 3. Revised SNIP indicator: For the journal-merger example, the revised SNIP gives the merged journal the same SNIP value as both original journals.The result follows from calculating the merged DCP exactly halfway between the two original DCP values.
4. Selection of citing journals
The revised SNIP excludes citing journals with unusually small or unreliable active-reference counts, addressing cases where such citations can disproportionately inflate a journal’s SNIP value. The selection procedure also applies trade-journal, continuity, and active-reference thresholds, while excluded journals retain SNIP values but may lose benefits from journal self-citations.
- Citations from trade journals and other special citing journals can disproportionately increase a cited journal’s SNIP value because their small numbers of active references reduce the DCP.This can make a citation from a trade journal or obscure national journal benefit SNIP more than a citation from a regular international scientific journal.
- The revised SNIP excludes journals with very small numbers of active references as citing journals.Citations given by excluded journals are not included in the revised SNIP calculation, although those journals still receive SNIP values.
- Excluded citing journals retain SNIP values but do not benefit from journal self-citations.This may disadvantage journals excluded from the citing-journal set.
- Journals classified as trade journals are excluded as citing journals in the first selection step.
- Journals must publish in the analysis year and each of the three preceding years to qualify as citing journals.The four-year continuity requirement aims to balance citing and cited publications in the revised SNIP calculation.
- At least 20% of a journal’s publications in the analysis year must have an active reference for the journal to qualify as a citing journal.An active reference points to a publication from one of the three preceding years in a journal included as a citing journal in the current year.
5. Empirical analysis
The empirical analysis compares revised SNIP values with the original indicator using 2010 data and journals meeting specified publication and citing-journal criteria. The revised indicator broadly tracks the original while producing lower average values and systematic field-related differences.
- Data and journal selection: 13,154 citing journals remained after excluding trade journals, journals lacking publications throughout 2007–2010, and journals with fewer than 20% of publications having active references.The selection began with 19,816 journals having at least one publication in 2010.
- Revised SNIP results: 1.03 was the publication-weighted average revised SNIP among citing journals, compared with 0.90 across all 22,434 journals with SNIP values.The citing-journal average was slightly above the theoretically expected value of one.
- Revised SNIP results: The revised SNIP ranking included journals from medicine, life sciences, natural sciences, computer science, engineering, and economics among its highest-ranked journals.The authors interpret this disciplinary diversity as an indication of correction for field differences.
- Revised SNIP results: RIP was 1.85 times as high as revised SNIP on average across 10,347 journals with at least 100 publications; life-science, medical, and chemistry journals tended to fall below this relation, while computer science, economics, and mathematics journals tended to lie above it.Some physics journals were also substantially above the average relation.
- Comparison with original SNIP: 26% higher was the publication-weighted average original SNIP than revised SNIP across 22,403 journals with values for both indicators.The respective averages were 1.13 for original SNIP and 0.90 for revised SNIP.
- Comparison with original SNIP: 0.93 was the Pearson correlation between original and revised SNIP values for 10,331 journals with at least 100 publications.Despite the strong overall relation, computer science and engineering journals appeared overrepresented among the highest original-SNIP rankings, and positive revised-minus-original differences were observed in medical, life, and natural sciences.
6. Conclusions
The revised SNIP preserves source normalization while addressing counterintuitive behaviors of the original indicator. Empirically, differences are small overall, but systematic ranking shifts and limitations remain.
- Source normalization corrects field differences without requiring explicitly defined field boundaries in a classification system.
- The original SNIP could decrease after an additional citation or fall below both component journals after a merger, whereas the revised indicator removes these properties.
- Empirical differences between the original and revised SNIP indicators are relatively small, but computer science and engineering journals tend to rank lower under the revised indicator.
- Average-based revised SNIP values are sensitive to highly cited outliers, which can substantially increase a journal’s score.
- Source normalization does not correct for differences in literature growth rates or unidirectional citation flows between fields.
Appendix A: Some properties of the revised SNIP indicator
The revised SNIP can be interpreted through weighted citation contributions and has more intuitive responses to citations and journal mergers. This perspective also clarifies its relationship to other source-normalized indicators.
- The revised SNIP can be reformulated mathematically to study its properties and compare it with other source-normalized indicators.
- The revised SNIP equals a journal’s average citations per publication, with each citation weighted by citing-publication references and citing-journal active-reference prevalence.
- The revised SNIP differs from the audience factor, fractional counting, and a priori normalization approaches in how citation-source characteristics are used for normalization.
- Each additional citation increases the cited journal’s revised SNIP, although the increase is smaller when the citing publication has more active references or the citing journal has higher active-reference prevalence.
- A merger of two journals produces a revised SNIP value between the original SNIP values of the two journals.
X XY SNIP SNIP SNIP m m
The revised SNIP of a merged journal is a weighted average of the SNIP values of the two original journals, so it lies between them.
- The merged journal’s SNIP is a weighted average of the SNIP values of journals X and Y.
- This weighted-average property guarantees that the merged journal’s SNIP lies between the two original journals’ SNIP values.
Appendix B: Source normalization in a formal mathematical framework
The appendix formalizes how the revised SNIP indicator corrects citation-practice differences between fields. Under three idealized assumptions, it shows that the publication-weighted average SNIP value equals one, while discussing the mechanism and limits of this result.
- Assumptions: The framework assumes no between-field citation traffic, stable annual publication output within each field, and at least one active-reference publication in every journal.The authors note that these assumptions do not hold perfectly in practice, so the mathematical results apply approximately depending on how closely reality matches the idealized framework.
- Formal framework: The analysis distinguishes cited and citing journals and publications, using three preceding cited years and one citing year.It defines N1 and N2 for cited and citing journals and m_k1 and m_k2 for their publication counts.
- Normalization criterion: The revised indicator evaluates the publication-weighted average SNIP value of cited journals within a field to assess whether field differences are properly corrected.The criterion is that the average μ has the same value across fields.
- Formal result: Under the three assumptions, the average SNIP value μ equals 1, demonstrating proper correction for field differences within the mathematical framework.The result follows by combining equation (B6) with the relation M1 = 3M2 implied by the three cited years and one citing year.
- Normalization mechanism: The revised SNIP corrects for citing publications without active references by increasing the value assigned to active references in publications that do have them.Without this adjustment, publications lacking active references provide no value to cited publications and distort field normalization.
- Sensitivity and scope: The appendix’s conclusions are sensitive to asymmetric cross-field citation, changing publication output, and journals lacking active references.Asymmetric citation can lower SNIP values in the citing field and raise them in the cited field; increasing output can raise the field average above one.