a global map of science based on the isi subject...
TRANSCRIPT
-
A Global Map of Science Based on the ISI SubjectCategories
Loet LeydesdorffAmsterdam School of Communications Research (ASCoR), University of Amsterdam, Kloveniersburgwal 48,1012 CX, Amsterdam, The Netherlands. E-mail: [email protected].
Ismael RafolsScience and Technology Policy Research (SPRU), University of Sussex, Freeman Centre, Falmer,Brighton, East Sussex BN1 9QE, United Kingdom. E-mail: [email protected].
The decomposition of scientific literature into disci-plinary and subdisciplinary structures is one of the coregoals of scientometrics. How can we achieve a gooddecomposition? The ISI subject categories classify jour-nals included in the Science Citation Index (SCI). Theaggregated journal-journal citation matrix contained inthe Journal Citation Reports can be aggregated on thebasis of these categories. This leads to an asymmetricalmatrix (citing versus cited) that is much more denselypopulated than the underlying matrix at the journal level.Exploratory factor analysis of the matrix of subject cate-gories suggests a 14-factor solution. This solution couldbe interpreted as the disciplinary structure of science.The nested maps of science (corresponding to 14 fac-tors, 172 categories, and 6,164 journals) are online athttp://www.leydesdorff.net/map06. Presumably, inaccu-racies in the attribution of journals to the ISI subjectcategories average out so that the factor analysis revealsthe main structures.The mapping of science could, there-fore, be comprehensive and reliable on a large scalealbeit imprecise in terms of the attribution of journals tothe ISI subject categories.
Introduction
The decomposition of the Science Citation Index intodisciplinary and subdisciplinary structures has fascinatedscientometricians and information analysts ever since thebeginning of this index. Price (1965) conjectured thatthe database would contain the structure of science. He sug-gested that journals would be the appropriate units of analysis,and that aggregated citation relations among journals mightreveal disciplinary and finer-grained delineations such asthose among specialties.
Received October 21, 2008; revised August 25, 2008; accepted August 25,2008
2008 ASIS&T Published online 3 October 2008 in Wiley InterScience(www.interscience.wiley.com). DOI: 10.1002/asi.20967
Carpenter and Narin (1973) tried to cluster the Sci-ence Citation Index database in terms of aggregated journalcitation patterns using the methods available at the time.However, the size of the database2,200 journals in 1969(Garfield, 1972, p. 472) and 6,164 journals in 2006makesit difficult to use algorithms more sophisticated than single-linkage clustering. Single-linkage clustering is based onrecursive selection of the two most-similar subsets, and usingrelational database management one can operate on rankorders in lists without loading the relatively large matricesinto memory.1
Small and Sweeney (1985) added a variable threshold tosingle-linkage clustering in their effort to map the sciencesglobally using cocitation analysis at the document level.However, the choice of thresholds, similarity criteria, andclustering algorithms was somewhat arbitrary. Because ofthe focus on relations, the latent dimensions of the matrix (itseigenvectors) could not be revealed using single-linkage clus-tering (Leydesdorff, 1987). A structural approach requiresmultivariate analysis, for example, based on distinguishingorthogonal dimensions using factor analysis: Units of analy-sis may be positioned similarly in a multidimensional spacewithout necessarily maintaining strong relations among them(Burt, 1982; Leydesdorff, 2006).
The factor-analytical approach is limited even today toapproximately 3,000 variables using the latest version ofSPSS,2 while in the meantime the Science Citation Indexhas grown to more than 6,000 journals. Most researchershave therefore focused on chunks of the database or used
1Single-linkage clustering (or nearest-neighbor) sorts relations hierarchi-cally in terms of decreasing order of their strength. The single strongest linkis clustered in each round. However, this may lead to chaining, whereelements cluster that otherwise might be considered as rather distant fromeach other.
2For technical reasons, SPSS can address approximately two gigabytes ofinternal memory of a computer as workspace.
JOURNAL OF THE AMERICAN SOCIETY FOR INFORMATION SCIENCE AND TECHNOLOGY, 60(2):348362, 2009
-
seed journals for collection of a sample (Doreian & Farraro,1985; Leydesdorff, 1986, 2006; Tijssen, de Leeuw, & vanRaan, 1987). In contrast, Boyack, Klavans, and Brner (2005)used the VxInsight algorithm (Davidson, Hendrickson,Johnson, Meyers, & Wylie, 1998) in order to map thewhole journal structure as a representation of the structureof science.3 Moya-Anagn et al. (2004, 2007) used cocita-tion and PathFinder for mapping the whole of science onthe basis of the ISI subject categories.4 However, subjectdelineations in the maps are based necessarily on trade-offsbetween accuracy, readability, and simplicity since the jour-nal sets are overlapping (Bensman, 2001; Boyack, Brner, &Klavans, 2007). Klavans and Boyack (2007, p. 438) notedthat a journal may occupy a different position in a differentcontext: Many journals report on developments in multipledisciplines; journals can also function as a major source ofreferences in more than one specialty. The position of eachjournal in the multidimensional space of journal-citation vec-tors allows for a specific perspective (Leydesdorff, 2007; Zitt,Ramanana-Rahary, & Bassecoulard, 2005).
In addition to interjournal citations, the Science CitationIndex database has been mapped using cocitations (Small,1973; Small & Griffith, 1974; Small & Sweeney, 1985) orco-occurrences of title words (Callon, Courtial, Turner, &Bauin, 1983; Callon, Law, & Rip, 1986; Leydesdorff, 1989),at various levels of aggregation. However, using lower-levelunits (like documents) instead of journals means abandon-ing Prices grandiose vision to map the whole of scienceusing the structure present in the aggregated journal-journal(co)citation matrix (Price, 1965).
Since citation relations among journals are dense indiscipline-specific clusters and otherwise virtually nonexis-tent, the journal-journal citation matrix can be considerednearly decomposable.5 While a decomposable matrix is asquare matrix such that an identical rearrangement of rowsand columns leaves a set of square submatrices on the prin-cipal diagonal and zeros everywhere else, in the case of anearly decomposable matrix some zeros are replaced by rela-tively small nonzero numbers (Simon &Ando, 1961;Ando &Fisher, 1963). Near-decomposability is a general property ofcomplex and evolving systems (Simon, 1973, 2002). Thenext-order units represented by the square submatricesand representing in this case disciplines or specialtiesarereproduced in relatively stable sets (of journals), whichmay change over time. The sets of journals are functionalsubsystems that show a high density in terms of relationswithin the center (i.e., core journals), but are more open tochange in relations at the margins. The organization amongthe subsystems can also change. The decomposition intonearly decomposable matrices has no analytical solution.However, algorithms can provide heuristic decompositions
3See http://mapofscience.com for maps of science based on this algorithm.4See http://www.atlasofscience.net for maps of science based on this
algorithm.5In 2006, the database contained only 1,201,562 of the 37,994,896
(= 6,1642) possible relations. This corresponds to a density of 3.16%.
when there is no single unique correct answer (Newman,2006a, 2006b).
ISI Subject Categories
Hitherto, the organization into components and clustershas been based mainly on the results of algorithms fromgraph or factor analysis operating on journal-journal citationmatrices. The designation is then based on an ex post factoappreciation of these results. However, the Institute of Scien-tific Information (ISI) has added a substantive classifier to thedatabase: the subject category or subject categories of eachjournal included. These categories are assigned by ISI staffon the basis of a number of criteria including the journalstitle and its citation patterns (Pudovkin & Garfield, 2002, atp. 1113; McVeigh, personal communication, March 9, 2006).
The subject categories of the ISI cannot be consideredas based on literary warrant like the classification of theLibrary of Congress (Chan, 1999). A classification schemebased on literary warrant is inductively developed in refer-ence to the holdings of a particular library, or to what is or hasbeen published (Leydesdorff & Bensman, 2006, p. 1473). Inother words, it is based on what the actual literature of thetime warrants. For example, each of the individual sched-ules in the classification of the Library of Congress (LC) wasinitially drafted by subject specialists, who consulted bib-liographies, treatises, comprehensive histories, and existingclassification schemes to determine the scope and content ofan individual class and its subclasses. The LC has a policyof continuous revision to take current literary warrant intoaccount, so that new areas are developed and obsolete ele-ments are removed or revised. The ISI categories, however,are changed in terms of respective coverage, but cannot berevised from the perspective of hindsight.
In order to enhance flexibility in the database, the Sci-ence Citation Index is organized with a CD-ROM versionfor each year separately (which is by definition fixed atthe date of delivery), and the SCI-Expanded version on theInternet, to which relevant data can be added from the per-spective of hindsight in order to optimize the database forinformation-retrieval purposes. The Journal Citation Reports,however, are provided as a separate service. The Web ver-sion of this database is kept in complete agreement withthe yearly CD-ROM. Thus, the subject categories themselvesare not systematically updated, although new categories canbe added and obsolete ones may no longer be used.
In addition to the subject categories, Thomson-ISI alsoassigns each journal in the Essential Science Indica-tors database (12,845 journals) to one of 22 so-called broadfields (at http://www.in-cites.com/journal-list/index.html)Journals are uniquely classified to a single broad field, whilethey can be classified under multiple subject categories inthe Science Citation Index. The Essential Science Indicatorsprovides statistics for government policy makers, universityor corporate research administrators, and so forth, while themain service of the Science Citation Index is informationretrieval for the research process.
JOURNAL OF THE AMERICAN SOCIETY FOR INFORMATION SCIENCE AND TECHNOLOGYFebruary 2009 349DOI: 10.1002/asi
-
FIG. 1. Frequency of 27 ISI Subject Categories that occur more than hundred times (JCR 2006).
Unlike the 22 broad fields, the 172+ ISI subject cate-gories are not hierarchically organized, but interconnected,because more than one category is often attributed to a jour-nal. Furthermore, they are more finely grained and thereforethis organization contains more information. In an attempt toreconstruct these subject categories on the basis of the aggre-gated journal-journal citation matrix, Leydesdorff (2006,p. 612) concluded that one cannot develop a conclusive clas-sification on the basis of a statistical analysis of citationrelations, but the quality of a proposed classification can betested against the structure in the data. Glnzel and Schubert(2003), for example, proposed 12 instead of 22 broad fields,but these categories are again different from the scheme of 12or 16 categories proposed by Boyack et al. (2005) and Moya-Anagn et al. (2007), respectively. In this study, we focuson the Science Citation Index, while these other studies alsoincluded the Social Science Citation Index. Given our factor-analytical approach, the differences in citation behavior andprocesses of codification between the social sciences and thenatural sciences could reduce the methods effectiveness byintroducing another source of variance (Leydesdorff, 2004;Leydesdorff & Hellsten, 2005). Bensman (2008), for exam-ple, argued that the impact factors of journals are differentlydistributed between the two databases.
The number of category attributions in the Science CitationIndex is 9,848 for 6,164 journals in 2006 or, in other words,approximately 1.6 categories per journal. The coverage ofthe 172 categories ranges from 262 journals sorted underBiochemistry and Molecular Biology to 5 journals sortedunder a single category.6 The average number of journals percategory is 56.3 (see Figure 1).
6Three more categories, which are no longer actively indexed, subsumeone or two journals.
The ISI subject categories match poorly with classifica-tions derived from the database itself on the basis of ananalysis of the principal components of the network matricesgenerated by citations (Leydesdorff, 2006, p. 611f). Using adifferent methodology, Boyack et al. (2005) found that insomewhat more than 50% of the cases the ISI categoriescorresponded closely with the clusters based on interjournalcitation relations. These results accord with the expectationthat many journals can be assigned unambiguous affiliationsin one core set or another, but the remainder, which is also alarge group, is heterogeneous (Garfield, 1971, 1972).
The ISI subject categories can be considered as macrojour-nals. Because more than one category can be attributed to ajournal, one can expect that the matrix of the citation relationsamong categories is less empty than the aggregated journal-journal citation matrix. However, the multidimensional spacespanned by these 172+ subject categories offers a wealth ofoptions for generating representations. One should not expecta unique map of science, but a number of possible repre-sentations (Leydesdorff, 2007; Zitt et al., 2005). Each mapcontains a projection from a specific perspective. However,one can ask whether there is a robust structure in terms of thelatent dimensions of the underlying matrix.
In other words, our research question is different fromBoyack et al.s (2005) effort to generate a new classificationusing a bottom-up strategy and from that of Moya-Anagnet al. (2007), who employed the ISI subject categories as unitsof measurement (at p. 2169), and used factor analysis of thecocitation matrix for the validation of their so-called factorscientograms. We wish to question the quality and valid-ity of using the ISI subject categories for mapping purposes.Can these subject classifications be used in further researchto demarcate the sciences and perhaps as field delineations,and if so, under what conditions? Like any classification,one can expect that these classifications can be used for
350 JOURNAL OF THE AMERICAN SOCIETY FOR INFORMATION SCIENCE AND TECHNOLOGYFebruary 2009DOI: 10.1002/asi
-
mapping purposes, but what is the value of these units oforganization?
This question is urgent because, first, there is a need for thedelineation of fields in citation analysis given that publica-tion and citation practices differ among fields of science (e.g.,Martin & Irvine, 1983; Moed, Burger, Frankfort, & van Raan,1985; Leydesdorff, 2008). Secondly, various studies of inter-disciplinarity have been based on the assumption that journalscan be grouped using the ISI subject categories (e.g., Morillo,Bordons, & Gmez, 2001, 2003; Van Leeuwen & Tijssen,2000). Interdisciplinarity is often a policy objective, whilenew developments may take place at borders of or across dis-ciplines (Zitt, 2005). One of the potential uses of a map ofscience is to help us understand the cognitive base and therelative positions of emerging fields (Bordons, Morillo, &Gmez, 2004; Porter, Cohen, Roessner, & Perreault, 2007;Porter, Roessner, Cohen, & Perreault, 2006;Van Raan, 2000).
As noted above, Moya-Anegn et al. (2007, p. 2173)used factor analysis of the cocitation matrix of the 218 cat-egories of the Science Citation Index and Social ScienceCitation Index (2002) combined for the validation of theirvisualizations. These authors stated that a scree test hadled them to the choice of 16 factors. The screeplot basedon Table 1 of their paper, is provided here in Figure 2.
TABLE 1. Highest factor loadings on the last factor in a 13-, 14-, and15-factor solution.
Highest factor loadings Highest factor loadings Highest factor loadingson factor 13 in the case on factor 14 in the case on factor 15 in the caseof a 13-factor solution of a 14-factor solution of a 15-factor solution
0.779 0.786 0.5930.715 0.721 0.5390.672 0.698 0.4840.669 0.687 0.472
FIG. 2. Screeplot of eigenvalues provided in Table 1 of Moya-Anegn et al. (2007, p. 2173).
In our opinion, this screeplot does not support the inferencebecause the line flattens after eight factors at the most. (Ascan be seen from Table 1 of Moya-Anegn et al.s paper,the 16-factor solution instead corresponds to including per-centages of variance explained equal or larger than unity.)Furthermore, this analysis was based on the symmetricalcocitation matrix instead of the asymmetrical citation matrix(cf. Leydesdorff & Vaughan, 2006). However, the focusof these authors was not on the analysis, but the visuali-zation technique (e.g., PathFinder) for showing relations andclusters; the factor analysis was successfully used to rational-ize the visualizations ex post facto. We approach the problemfirst factor-analytically using the asymmetrical matrix ofaggregated citations among categories, and will subsequentlytry to map the sciences hierarchically top-down insofar as ourresults show that it is legitimate for us to do so.
Methods
The data was harvested from the CD-ROM version ofthe Journal Citation Reports of the Science Citation Index2006. As indicated above, 175 subject categories are used.Three categories (Psychology, biological, Psychology,experimental, and Transportation) are no longer used asclassifiers in the citing dimension, but four journals are stillindicated with these three categories in the cited dimension.Thus, we work with 172 citing and 175 cited categories.
The matrix, accordingly, contains two structures: a citedand a citing one. Saltons cosine was used for normalizationin both the cited and citing directions (Ahlgren, Jarneving, &Rousseau, 2003; Salton & McGill, 1983). The cosine is equalto the Pearson correlation, but without normalization to thearithmetic mean (Jones & Furnas, 1987). Pajek is used forthe visualizations (Batagelj & Mrvar, 2007) and SPSS (v15)for the factor analysis. The threshold for the visualizations is
JOURNAL OF THE AMERICAN SOCIETY FOR INFORMATION SCIENCE AND TECHNOLOGYFebruary 2009 351DOI: 10.1002/asi
-
pragmatically set at cosine 0.2.7 Visualizations are basedon the algorithm of Kamada and Kawai (1989). The size ofthe nodes is proportional to the number of citations in a givencategory (or in Figure 4, below, the logarithm of this value).The thickness and grey-shade of the links is proportionalto the cosine values in five equal steps of 0.2. The threshold ofcosine 0.2 will consistently be used for the visualizationsat the various levels.8
Using a factor model, the crucial question is the numberof factors to be extracted. Unless one has a priori reasons fortesting an assumption, this number has to be determined onempirical grounds. Unlike principal-component analysis, therotated component matrix is not an analytical rewrite ofthe original data. While both principal-component analysisand rotated component analysis allow for data reduction, thecriterion for the optimization in the case of factor analysisis no longer to explain as much variance as possible in thedata, but to find common factors in the set that explainthe covariance between the variables. Factor analysis is nec-essarily based on an assumption about the number of factorsthat span the multidimensional space (Leydesdorff, 2006).SPSS includes by default all factors with an eigenvalue largerthan unity. However, the resulting screeplot of the eigenval-ues can be used for an initial assessment of the number ofmeaningful factors. This assumption has to be tested againstthe data (Kim, 1975; Kim & Mueller, 1978).
Results
Unlike the aggregated journal-journal citation matrix, thematrix of 172 (citing) times 175 (cited) categories is notsparse: 11,577 of the (172 175=) 30,100 cells have a zerovalue. This corresponds to 38.46% of the number of cells.9
Since the categories are unevenly distributed, one cannot seta threshold value across the matrix without normalization.The factor analysis, however, begins with a normalizationusing the Pearson correlation coefficient. As noted, the visu-alizations are based on cosine values (Egghe & Leydesdorff,2008).
Let us focus on the structure in the citing dimensionbecause this structure is actively maintained by the indexingservice and is therefore current. The screeplot of the eigen-values suggests further exploration of a 14-factor solutionbecause the continuity in the curve is interrupted at the valueof 14 in Figure 3.
7A threshold is needed for the visualization because the cosine-basednetworks are often almost complete. Using the cosine, a threshold cannot beset on analytical grounds.
8The effects of a threshold at cosine 0.2 on the density of the matricesunderlying the figures in this study, are as follows:
cosine 0.0 cosine 0.2Figure 4 0.994 (N = 172) 0.175 (N = 171)Figure 5 0.929 (N = 14) 0.357 (N = 14)
9UCINet computes a density for this matrix after binarization of0.7538 0.4308.
FIG. 3. Scree plot of the factor analysis (citing).
Table 1 shows the four highest loadings on the last factorin the case of extracting 13, 14, or 15 factors in the citingdimension, respectively. This confirms that the quality ofthe factors declines considerably after extracting 14 factors.The 14-factor solution explains 51.8% of the variance of thematrix in the citing projection (and 47.9% of the variance inthe cited projection).
The factor loadings for the 172 categories on the 14 fac-tors in the citing dimension are provided in the Appendix.They can be interpreted in terms of disciplines, such asphysics, chemistry, clinical medicine, neurosciences, engi-neering, and ecology. (These designations are ours.) Thefactors in the cited dimension can be designated using pre-cisely the same disciplinary classifications, but their rankorder (that is, the percentage of variance explained by eachfactor) is different (Table 2). Out of the 172 categories, 154(89%) fall in the same factor in both the citing and cited pro-jections. The 18 categories that are classified differently inthe citing and cited projections are listed in Table 3.
TABLE 2. Fourteen disciplines distinguished on the basis of ISI subjectcategories in 2006 ( > .95; p < .01).
Citing factors Cited factors
Biomedical sciences 1 1Materials sciences 2 2Computer sciences 3 4Clinical medicine 4 5Neurosciences 5 3Ecology 6 7Chemistry 7 9Geosciences 8 6Engineering 9 8Infectious diseases 10 10Environmental sciences 11 12Agriculture 12 11Physics 13 13General medicine; health 14 14
352 JOURNAL OF THE AMERICAN SOCIETY FOR INFORMATION SCIENCE AND TECHNOLOGYFebruary 2009DOI: 10.1002/asi
-
TABLE 3. Eighteen ISI subject categories that are classified differently in the citing and cited dimensions.
ISI subject category Citing factor Cited factor
Urology & nephrology Biomedical sciences Clinical medicinePharmacology & pharmacy Biomedical sciences NeurosciencesPhysiology Biomedical sciences NeurosciencesMedicine, legal Biomedical sciences ChemistryToxicology Biomedical sciences Environmental sciencesBiotechnology & applied microbiology Biomedical sciences AgricultureNutrition & dietetics Biomedical sciences AgricultureMathematical & computational biology Biomedical sciences General medicine; healthEnergy & fuels Materials sciences EngineeringComputer science, Interdisciplinary Applications Computer sciences EngineeringMathematics Engineering Computer sciencesEngineering, industrial Computer sciences PhysicsChemistry, physical Chemistry Materials sciencesMaterials science, biomaterials Chemistry Materials sciencesChemistry, applied Chemistry AgricultureMaterials science, composites Engineering Materials sciencesMycology Infectious diseases AgricultureMedicine, general & internal Medicine, general Clinical medicine
The strong overlap between the results of the factor anal-ysis in the cited and the citing dimension (Table 2) suggeststhat the matrix is nearly decomposable in terms of centraltendencies. Table 3 indicates cases where the scholars pub-lishing in one category cite on average differently from howthey are cited. For example, Mathematics exhibits a nega-tive factor loading on the engineering dimension in its citingpattern, while Mathematics, applied is loading primarilyon this dimension. In the cited dimension, however, the twocategories are both classified as engineering. This relativeinterdisciplinarity of mathematics accords with the find-ings by the SCImago Group (Moya-Anegn et al., 2007,p. 2172). Using a different technique (see above), they couldalso not find a separate factor for mathematics.10
In other cases, it is more difficult to provide an interpre-tation of the differences. Why would Biotech and AppliedChemistry be assigned to Agriculture in the cited dimensionand to Biomedical Sciences in the citing dimension? Is thisan indication of the interdisciplinary relation between thesetwo contexts of application for biotechnology?
Figure 4 shows the map of 171 ISI subject categories thatrelate above the threshold of cosine 0.2.11 (Only the cat-egory Agricultural Economics and Politics is no longerrelated at this threshold level.) The nodes represent the cate-gories and are colored in terms of the 14 factors. (The picture
10In a response to the critique of the International Mathematical Unionon the use of citation analysis (Adler, Ewing, & Taylor 2008), Bensman(June 27, 2008, at http://listserv.utk.edu/cgi-bin/wa?A2=ind0806&L=sigmetrics&D=1&O=D&P=16332) noted that the range of impact factorsamong mathematics journals was extraordinarily low and tight and the topjournals on the impact factor had no review articles. He added, This issuggestive of an extremely random citation pattern with no developmentof consensual paradigms. Therefore, math acts like a humanities in terms ofits literature use, and citation analysis is probably not applicable to thisdiscipline.
11A colored version of this map can be retrieved at http://www.leydesdorff.net/map06/Figure4.
in the cited dimension is very similar.) In this chart, the nodesizes were set proportional to the logarithm of the number ofcitations (in the respective subject category) in order to keepthe visualization readable.
Whereas the traditional disciplines are represented byclear factors (e.g., Physics or Chemistry), specific fields ofapplication in mathematics or engineering do not fall inthe disciplinary classification, but in the factor representingtheir topic. For example, Mathematical Physics is classi-fied as Physics. However, Chemical Engineering loads onthe Chemistry factor more than on the one representingengineering.
Figure 5 shows the citation relations among the 14 groups.(The depiction in the cited dimension is again very sim-ilar to this one in the citing dimension.) While Figure 4gave greater detail about the relations among subdisciplinesand specialties, the factor-analytical categories allow us todepict these ISI subject categories in Figure 4 with differ-ent colors in terms of the disciplinary affiliations provided inFigure 5. Both levels are interactively related with hyperlinksat http://www.leydesdorff.net/map06/index.htm.
The largest factor is designated as Biomedical Sciences.It includes at the disaggregated level
1. The core biological sciences, such as Biochemistry andMolecular Biology, Developmental Biology, Genetics,and Cell Biology;
2. The methodologies crucial for the biological sciences,such as Microscopy and Biochemical Research Methods;
3. Disciplines that fall into medicine but are highly interre-lated with the biological sciences, such as Oncology andPathology.
Among the latter, eight subject categories (e.g., Physiol-ogy, Toxicology, or Nutrition Sciences) have a citing patternin the factor of the Biomedical Sciences (that is, they drawon basic biological knowledge), but they show a cited pattern
JOURNAL OF THE AMERICAN SOCIETY FOR INFORMATION SCIENCE AND TECHNOLOGYFebruary 2009 353DOI: 10.1002/asi
-
FIG. 4. 171 ISI subject categories in the citing dimension; cosine 0.2. Node sizes set proportional to the logarithm of the number of citations given byeach category. A colored version of this map can be retrieved at http://www.leydesdorff.net/map06/Figure4.
FIG. 5. Fourteen disciplines in the citing dimension; cosine 0.2. (The colors correspond with those used in Figure 4.) A colored version of this map canbe retrieved at http://www.leydesdorff.net/map06.
354 JOURNAL OF THE AMERICAN SOCIETY FOR INFORMATION SCIENCE AND TECHNOLOGYFebruary 2009DOI: 10.1002/asi
-
in factors more related to specific applications, such as Neu-rosciences, Environmental Sciences, or Infectious Diseases(see Table 3 above).
Four factors are closely associated with the BiomedicalSciences: Clinical Medicine, Neurosciences, Infectious Dis-eases, and General Medicine and Health.At the opposite poleof the medicine-related factors, we find factors based on thehard sciences: the factors of Physics, Engineering, MaterialsSciences, and Computer Sciences, are among them. Chem-istry plays a brokerage role between Physics and MaterialSciences, on the one side, and core Biomedical Sciences suchas Biophysics and Biochemistry, on the other.
The relative positions of the subject categories withinFigure 4 inform us prima facie about their disciplinary orinterdisciplinary character. However, one should be cautiousin drawing conclusions from a visual inspection of maps. Amap remains a two-dimensional projection of a space (in thiscase, a 14-dimensional one), and one therefore needs a largenumber of projections from different angles before one canformulate hypotheses on this basis. On the basis of a numberof these projectionsthat is, variants of Figure 4we feelcomfortable in suggesting that the connection between themedical pole and the hard-science pole is achieved byway of three main routes:
1. A direct link between the Computer Sciences and some ofthe medical specialties such as Psychology, Neuroimag-ing, and Medical Informatics;
2. Through Chemistry, which appears to play a brokeragerole between Physics and Material Sciences, on the oneside, and the core Biomedical Sciences such as Biophysicsand Biochemistry, on the other;
3. Through a path that links Engineering and Material Sci-ences with Geosciences and Environmental Sciences, andalso connects these two latter factors with Ecology andAgriculture. The latter are related to Infectious Diseasesand the large set of journals in the Biomedical Sciences.This path can be considered as a small cluster with a focuson environmental issues.
Our results are consistent with previously reported maps(Boyack et al. 2005; Boyack & Klavans, 2007; Moya-Anagnet al., 2007), but we chose to exclude the social sciences.We would expect differences and similarities when map-ping the social sciences (using the Social Science CitationIndex) because of the different order of magnitude of cita-tions in the journal-journal citation network, differences incitation behavior and codification processes (Leydesdorff &Hellsten, 2005), and the potentially different functionsof citations as relations among texts in these sciences(Bensman, 2008).
Conclusions and Discussion
Why do the ISI subject categories that were found to bea poor match for journal-citation patterns in other research(Boyack et al., 2005; Leydesdorff, 2006) perform relativelywell when aggregated in order to provide comprehensive
maps of science in both the cited and citing dimensions?The explanation is statistical: Boyack et al. (2005) notedthat the ISI subject categories match in approximately 50%of the cases and mismatch consequently in the remaining50%. The error, however, is not systematic so that the 50%matching cases prevail in the aggregate. Factor analysisenables us to distinguish the pattern as a signal from thenoise. Thus, a clear factor structure can be discerned at thisintermediate level.
From the top-down perspective of the factor structure,the noise at the bottom level can be considered as merevariation, which is distributed stochastically. Factor analy-sis enables us to reduce the complexity in the data. As wenoted above, the resulting maps match well with the previ-ously published mappings of the team of Boyack, Brner, andKlavans and the ones of the SCImago Group (see http://www.atlasofscience.net and http://mapofscience.com, respec-tively). The matrix of aggregated intercategory citationsused in this study is available at http://www.leydesdorff.net/map06/data.xls for users to draw their own maps or maketheir own extractions and inferences.
The maps are also available as a nested structure athttp://www.leydesdorff.net/map06/index.htm. One can clickon each of the 14 categories visualized in Figure 5, opena map of the corresponding discipline in terms of subjectcategories, and then relate to the journal sets subsumedunder the respective category. Top-down one is thus guidedto the individual journal maps as available at http://www.leydesdorff.net/jcr06/citing/index.htm. Like the other maps,our maps have the disadvantage of being static representa-tions of science, based on a single year of data. In futureresearch, we hope to expand this line of research withdynamic analysis of journal maps (Leydesdorff & Schank,2008). Systematic comparisons between maps based on theScience Citation Index and Social Science Citation Indexremain also part of our research agenda.
Acknowledgment
We are grateful for suggestions and remarks of a numberof anonymous referees.
References
Adler, R., Ewing, J., & Taylor, P. (2008). Citation statistics: A report fromthe International Mathematical Union (IMU) in cooperation with theInternational Council of Industrial and Applied Mathematics (ICIAM)and the Institute of Mathematical Statistics (IMS). Retrieved fromhttp://www.mathunion.org/fileadmin/IMU/Report/CitationStatistics.pdf
Ahlgren, P., Jarneving, B., & Rousseau, R. (2003). Requirement for a Cocita-tion Similarity Measure, with Special Reference to Pearsons CorrelationCoefficient. Journal of the American Society for Information Science andTechnology, 54, 550560.
Ando, A., & Fisher, F.M. (1963). Near-decomposability, partition and aggre-gation, and the relevance of stability discussions. International EconomicReview, 4(1), 5367.
Batagelj, V., & Mrvar, A. (2007) Pajek. Program for large networkanalysis. Retrieved on October 4, 2007, from http://vlado.fmf.uni-lj.si/pub/networks/pajek/
JOURNAL OF THE AMERICAN SOCIETY FOR INFORMATION SCIENCE AND TECHNOLOGYFebruary 2009 355DOI: 10.1002/asi
-
Bensman, S.J. (2001). Bradfords law and fuzzy sets: Statistical implicationsfor library analyses. IFLA Journal, 27, 238246.
Bensman, S.J. (2008). Distributional differences of the impact factor in thesciences versus the social sciences: An analysis of the probabilistic struc-ture of the 2005 Journal Citation Reports. Journal of theAmerican Societyfor Information Science and Technology, 59, 13661382.
Bordons, M., Morillo, F., & Gmez, I. (2004). Analysis of cross-disciplinaryresearch through bibliometric tools. In H.F. Moed, W. Glnzel &U. Schmoch (Eds.), Handbook of quantitative science and technologyresearch (pp. 437456). Dordrecht, Netherlands: Kluwer.
Boyack, K., Brner, K., & Klavans, R. (2007). Mapping the structure andevolution of chemistry research. In D. Torres-Salinas & H. Moed (Eds.),Proceedings of the Eleventh International Conference on Scientometricsand Informetrics (Vol. 1, pp. 112123). Madrid, Spain: CSIC.
Boyack, K., & Klavans, R. (2007).A map of science. Retrieved on October 4,2007, from http://mapofscience.com/
Boyack, K.W., Klavans, R., & Brner, K. (2005). Mapping the backbone ofscience. Scientometrics, 64, 351374.
Burt, R.S. (1982). Toward a structural theory of action. NewYork: AcademicPress.
Callon, M., Courtial, J.-P., Turner, W.A., & Bauin, S. (1983). From transla-tions to problematic networks: An introduction to coword analysis. SocialScience Information, 22, 191235.
Callon, M., Law, J., & Rip, A. (Eds.). (1986). Mapping the dynamics ofscience and technology. London: Macmillan.
Carpenter, M.P., & Narin, F. (1973). Clustering of scientific journals. Journalof the American Society for Information Science, 24, 425436.
Chan, L.M. (1999). A guide to the Library of Congress classification (5thed). Englewood, CO: Libraries Unlimited.
Davidson, G.S., Hendrickson, B., Johnson, D.K., Meyers, C.E., &Wylie, B.N. (1998). Knowledge mining with VxInsight: Discoverythrough interaction. Journal of Intelligent Information Systems, 11,259285.
Doreian, P., & Fararo, T.J. (1985). Structural equivalence in a journal net-work. Journal of the American Society for Information Science, 36,2837.
Egghe, L., & Leydesdorff, L. (2008). The relation between Pearsonscorrelation coefficient r and Saltons cosine measure. Manuscript inpreparation.
Garfield, E. (1971). The mystery of the transposed journal lists, whereinBradfords law of scattering is generalized according to Garfields law ofconcentration. Current Contents, 3(33), 56.
Garfield, E. (1972). Citation analysis as a tool in journal evaluation. Science,178(4060), 471479.
Glnzel, W., & Schubert, A. (2003). A new classification scheme of sci-ence fields and subfields designed for scientometric evaluation purposes.Scientometrics, 56, 357367.
Jones, W.P., & Furnas, G.W. (1987). Pictures of relevance: A geometricanalysis of similarity measures. Journal of the American Society forInformation Science, 36, 420442.
Kamada, T., & Kawai, S. (1989). An algorithm for drawing generalundirected graphs. Information Processing Letters, 31(1), 715.
Kim, J.-O. (1975). Factor analysis. In N.H. Nie, C.H. Hull, J.G. Jenkins,K. Steinbrenner, & D.H. Bent (Eds.), SPSS: Statistical package for thesocial sciences (pp. 468514). New York: McGraw-Hill.
Kim, J.-O., & Mueller, C.W. (1978). Factor analysis, statistical methods, andpractical issues. Beverly Hills, CA: Sage.
Klavans, R., & Boyack, K. (2007). Is there a convergent structure of sci-ence? A comparison of maps using the ISI and Scopus databases. InD. Torres-Salinas & H. Moed (Eds.), Proceedings of the EleventhInternational Conference of Scientometrics and Informetrics (Vol. 1,pp. 437448).
Leydesdorff, L. (1986). The development of frames of references. Sciento-metrics, 9, 103125.
Leydesdorff, L. (1987). Various methods for the mapping of science.Scientometrics, 11, 291320.
Leydesdorff, L. (1989). Words and cowords as indicators of intellectualorganization. Research Policy, 18, 209223.
Leydesdorff, L. (2004). Top-down decomposition of the Journal CitationReport of the Social Science Citation Index: Graph- and factor-analyticalapproaches. Scientometrics, 60, 159180.
Leydesdorff, L. (2006). Can scientific journals be classified in terms of aggre-gated journal-journal citation relations using the Journal Citation Reports?Journal of the American Society for Information Science and Technology,57, 601613.
Leydesdorff, L. (2007). Visualization of the citation impact environments ofscientific journals: An online mapping exercise. Journal of the AmericanSociety of Information Science and Technology, 58, 207222.
Leydesdorff, L. (2008). Caveats for the use of citation indicators in researchand journal evaluation. Journal of the American Society for InformationScience and Technology, 59, 278287.
Leydesdorff, L., & Bensman, S.J. (2006). Classification and powerlaws:The logarithmic transformation. Journal of the American Society forInformation Science and Technology, 57, 14701486.
Leydesdorff, L., & Hellsten, I. (2005). Metaphors and diaphors in sci-ence communication: Mapping the case of stem-cell research. ScienceCommunication, 27(1), 6499.
Leydesdorff, L., & Schank, T. (2008). Dynamic animations of journal maps:Indicators of structural change and interdisciplinary developments. Jour-nal of the American Society for Information Science and Technology, 59,18101818.
Leydesdorff, L., & Vaughan, L. (2006). Co-occurrence matrices and theirapplications in information science: Extending ACA to the Web envi-ronment. Journal of the American Society for Information Science andTechnology, 57, 16161628.
Martin, B., & Irvine, J. (1983). Assessing basic research: Some partial indi-cators of scientific progress in radio astronomy. Research Policy, 12,6190.
Moed, H.F., Burger, W.J.M., Frankfort, J.G., & van Raan, A.F.J. (1985).The use of bibliometric data for the measurement of university researchperformance. Research Policy, 14, 131149.
Morillo, F., Bordons, M., & Gmez, I. (2001). An approach tointerdisciplinarity through bibliometric indicators. Scientometrics, 51,203222.
Morillo, F., Bordons, M., & Gmez, I. (2003). Interdisciplinarity in sci-ence: A tentative typology of disciplines and research areas. Journalof the American Society for Information Science and Technology, 54,12371249.
Moya-Anegn, F., Vargas-Quesada, B., Chinchilla-Rodrguez, Z., Corera-lvarez, E., Munoz-Fernndez, F.J., & Herrero-Solana, V. (2007). Visu-alizing the marrow of science. Journal of the American Society forInformation Science and Technology, 58, 21672179.
Moya-Anegn, F., Vargas-Quesada, B., Herrero-Solana, V., Chinchilla-Rodrguez, Z., Corera-lvarez, E., & Munoz-Fernndez, F.J. (2004).A new technique for building maps of large scientific domains based onthe cocitation of classes and categories. Scientometrics, 61, 129145.
Newman, M.E.J. (2006a). Finding community structure in networks usingthe eigenvectors of matrices. Physical Review E, 74, 36104.
Newman, M.E.J. (2006b). Modularity and community structure in net-works. Proceedings of the National Academy of Sciences, USA, 103(23),85778582.
Porter,A.L., Cohen,A.S., Roessner, J.D., & Perreault, M. (2007). Measuringresearcher interdisciplinarity. Scientometrics, 72, 117147.
Porter, A.L., Roessner, J.D., Cohen, A.S., & Perreault, M. (2006). Inter-disciplinary research: Meaning, metrics, and nurture. Research Evalua-tion, 15(3), 187195.
Price, D.J. de Solla (1965). Networks of scientific papers. Science,149(3683), 510515.
Pudovkin, A.I., & Garfield, E. (2002). Algorithmic procedure for find-ing semantically related journals. Journal of the American Society forInformation Science and Technology, 53(13), 11131119.
Salton, G., & McGill, M.J. (1983). Introduction to modern informationretrieval. Auckland, New Zealand: McGraw-Hill.
Simon, H.A. (1973). The organization of complex systems. In H.H. Pattee(Ed.), Hierarchy theory: The challenge of complex systems (pp. 127).New York: George Braziller.
356 JOURNAL OF THE AMERICAN SOCIETY FOR INFORMATION SCIENCE AND TECHNOLOGYFebruary 2009DOI: 10.1002/asi
-
Simon, H.A. (2002). Near decomposability and the speed of evolution.Industrial and Corporate Change, 11, 587599.
Simon, H.A., & Ando, A. (1961). Aggregation of variables in dynamicsystems. Econometrica, 29, 111138.
Small, H. (1973). Cocitation in the scientific literature: A new measure ofthe relationship between two documents. Journal of the American Societyfor Information Science, 24, 265269.
Small, H., & Griffith, B. (1974). The structure of scientific literature I.Science Studies, 4, 1740.
Small, H., & Sweeney, E. (1985). Clustering the science citation index usingcocitations I: A comparison of methods. Scientometrics, 7, 391409.
Tijssen, R., de Leeuw, J., & van Raan, A.F.J. (1987). Quasi-correspondenceanalysis on square scientometric transaction matrices. Scientometrics, 11,347361.
Van Leeuwen, T., & Tijssen, R. (2000). Interdisciplinary dynamics ofmodern science: Analysis of cross-disciplinary citation flows. ResearchEvaluation, 9(3), 183187.
Van Raan, A.F.J. (2000). The interdisciplinary nature of science: Theoret-ical framework and bibliometric-empirical approach. In P. Weingart &N. Stehr (Eds.), Practicing interdisciplinarity (pp. 6678). Toronto:University of Toronto Press.
Zitt, M. (2005). Facing diversity of science: A challenge for bibliometricindicators. measurement: Interdisciplinary research and perspectives,3(1), 3849.
Zitt, M., Ramanana-Rahary, S., & Bassecoulard, E. (2005). Relativityof citation performance and excellence measures: From cross-field to cross-scale effects of field-normalisation. Scientometrics, 63,373401.
JOURNAL OF THE AMERICAN SOCIETY FOR INFORMATION SCIENCE AND TECHNOLOGYFebruary 2009 357DOI: 10.1002/asi
-
Ap
pen
dix
:Fac
tor
An
alys
iso
f17
2IS
ISu
bje
ctC
ateg
ori
esO
ver
the
Cit
ing
Pro
ject
ion
Citi
ngfa
ctor
ISI
Subj
ectC
ateg
ory
Biomedicalsciences
Materialssciences
Computersciences
Clinicalsciences
Neurosciences
Ecology
Chemistry
Geosciences
Engineering
InfectiousDiseases
Environmentalsciences
Agriculture
Physics
Gen.medicine;health
Bio
med
ical
scie
nces
Cel
lbio
logy
0.89
00.
118
0.19
2B
ioch
emis
try
&m
olec
ular
biol
ogy
0.85
70.
212
0.21
50.
170
Bio
phys
ics
0.81
80.
270
0.17
10.
136
Dev
elop
men
talb
iolo
gy0.
818
Mul
tidis
cipl
inar
ysc
ienc
es0.
802
0.18
50.
209
0.14
50.
167
0.24
30.
128
0.12
6G
enet
ics
&he
redi
ty0.
773
0.25
40.
149
0.22
2B
iolo
gy0.
746
0.24
10.
439
0.10
40.
107
Med
icin
e,re
sear
ch&
expe
rim
enta
l0.
738
0.35
10.
113
0.43
50.
113
Mic
rosc
opy
0.71
80.
283
0.1
14A
nato
my
&m
orph
olog
y0.
711
0.13
40.
261
0.12
70
.143
End
ocri
nolo
gy&
met
abol
ism
0.64
50.
151
0.16
20.
166
Bio
tech
nolo
gy&
appl
ied
mic
robi
olog
y0.
622
0.20
40.
367
0.37
6Ph
ysio
logy
0.60
10.
254
0.43
2R
epro
duct
ive
biol
ogy
0.59
40
.115
0.1
990
.110
0.1
420
.145
0.1
160.
147
And
rolo
gy0.
592
0.1
030
.105
0.1
870
.114
0.1
530
.187
0.12
80
.124
0.20
5M
edic
alla
bora
tory
tech
nolo
gy0.
586
0.41
20.
199
0.24
1B
ioch
emic
alre
sear
chm
etho
ds0.
575
0.38
20.
187
0.20
4Pa
thol
ogy
0.56
00.
369
0.12
70.
204
Onc
olog
y0.
559
0.24
80.
151
Mat
hem
atic
al&
com
puta
tiona
lbio
logy
0.54
30
.162
0.17
30.
254
0.22
90.
166
0.1
890.
221
Toxi
colo
gy0.
511
0.14
20.
182
0.17
30.
339
0.11
20.
175
phar
mac
olog
y&
phar
mac
y0.
500
0.16
30.
429
0.25
10.
275
0.10
90.
149
Obs
tetr
ics
&gy
neco
logy
0.40
00
.130
0.2
140
.107
0.1
300
.199
0.11
90
.138
0.31
9N
utri
tion
&di
etet
ics
0.36
90.
109
0.1
210.
236
0.25
8M
edic
ine,
lega
l0.
346
0.11
70.
164
0.14
90.
180
Uro
logy
&ne
phro
logy
0.24
10.
241
0.15
2
Mat
eria
lsci
ence
sM
ater
ials
scie
nce,
mul
tidis
cipl
inar
y0.
913
0.23
90.
161
Nan
osci
ence
&na
note
chno
logy
0.87
80.
301
0.11
4M
ater
ials
scie
nce,
coat
ings
&fil
ms
0.86
00.
107
Phys
ics,
appl
ied
0.82
80.
151
0.25
0M
ater
ials
scie
nce,
cera
mic
s0.
785
0.13
3M
etal
lurg
y&
met
allu
rgic
alen
gine
erin
g0.
771
0.15
4Ph
ysic
s,co
nden
sed
mat
ter
0.72
10.
136
0.37
1M
ater
ials
scie
nce,
char
acte
riza
tion
&0.
676
0.48
3te
stin
gM
inin
g&
min
eral
proc
essi
ng0.
557
0.27
30.
112
Inst
rum
ents
&in
stru
men
tatio
n0.
518
0.29
50.
102
0.35
4E
lect
roch
emis
try
0.46
70.
295
Ene
rgy
&fu
els
0.30
10.
234
0.27
80.
239
358 JOURNAL OF THE AMERICAN SOCIETY FOR INFORMATION SCIENCE AND TECHNOLOGYFebruary 2009DOI: 10.1002/asi
-
Com
pute
rsc
ienc
esC
ompu
ter
scie
nce,
hard
war
e&
0.88
2ar
chite
ctur
eC
ompu
ter
scie
nce,
info
rmat
ion
syst
ems
0.81
4C
ompu
ter
scie
nce,
artifi
cial
inte
llige
nce
0.81
2E
ngin
eeri
ng,e
lect
rica
l&el
ectr
onic
0.33
50.
793
0.12
3C
ompu
ter
scie
nce,
The
ory
&m
etho
ds0.
791
Com
pute
rsc
ienc
e,so
ftw
are
engi
neer
ing
0.75
0Te
leco
mm
unic
atio
ns0.
114
0.74
40
.109
0.10
4C
ompu
ter
scie
nce,
cybe
rnet
ics
0.72
90.
330
Aut
omat
ion
&co
ntro
lsys
tem
s0.
718
Tra
nspo
rtat
ion
scie
nce
&te
chno
logy
0.55
30
.110
0.18
90.
118
Com
pute
rsc
ienc
e,in
terd
isci
plin
ary
0.27
10
.133
0.51
40.
317
0.43
90.
112
0.11
3ap
plic
atio
nsR
obot
ics
0.48
4O
pera
tions
rese
arch
&m
anag
emen
t0.
420
0.20
10
.154
0.1
78sc
ienc
eM
athe
mat
ics,
appl
ied
0.1
590.
329
0.29
10
.159
0.19
3E
ngin
eeri
ng,i
ndus
tria
l0.
314
0.25
60
.148
0.2
48
Clin
ical
med
icin
eSu
rger
y0.
796
Cri
tical
care
med
icin
e0.
706
0.12
30.
131
0.14
8E
mer
genc
ym
edic
ine
0.66
60.
283
Tra
nspl
anta
tion
0.11
80.
636
0.30
4R
espi
rato
rysy
stem
0.12
00.
617
0.16
70.
126
Peri
pher
alva
scul
ardi
seas
e0.
285
0.61
60.
173
Car
diac
&ca
rdio
vasc
ular
syst
ems
0.14
60.
608
0.15
1O
rtho
pedi
cs0.
551
0.16
0E
ngin
eeri
ng,b
iom
edic
al0.
196
0.54
60.
146
0.1
430
.116
Hem
atol
ogy
0.46
10.
490
0.18
30.
132
Pedi
atri
cs0.
445
0.17
40
.103
0.11
20.
293
Spor
tsci
ence
s0.
381
0.27
10
.117
Ane
sthe
siol
ogy
0.37
40.
295
Gas
troe
nter
olog
y&
hepa
tolo
gy0.
137
0.34
30.
166
Rad
iolo
gy,n
ucle
arm
edic
ine
&0.
120
0.33
50.
257
0.1
50m
edic
alim
agin
gO
torh
inol
aryn
golo
gy0.
328
0.16
1R
heum
atol
ogy
0.12
00.
217
0.16
30.
141
Der
mat
olog
y0.
178
0.20
70.
205
Den
tistr
y,or
alsu
rger
y&
med
icin
e0.
197
(Con
tinu
ed)
JOURNAL OF THE AMERICAN SOCIETY FOR INFORMATION SCIENCE AND TECHNOLOGYFebruary 2009 359DOI: 10.1002/asi
-
Ap
pen
dix
.(C
onti
nued
)
Citi
ngfa
ctor
ISI
Subj
ectC
ateg
ory
Biomedicalsciences
Materialssciences
Computersciences
Clinicalsciences
Neurosciences
Ecology
Chemistry
Geosciences
Engineering
InfectiousDiseases
Environmentalsciences
Agriculture
Physics
Gen.medicine;health
Neu
rosc
ienc
esN
euro
scie
nces
0.27
10.
870
Psyc
holo
gy0.
807
Beh
avio
rals
cien
ces
0.16
40.
804
0.25
5N
euro
imag
ing
0.19
80.
792
0.1
04Ps
ychi
atry
0.78
30.
174
Clin
ical
neur
olog
y0.
310
0.75
2Su
bsta
nce
abus
e0.
641
0.14
20.
222
Ger
iatr
ics
&ge
ront
olog
y0.
395
0.19
40.
617
0.27
4R
ehab
ilita
tion
0.37
90.
516
0.1
06O
phth
alm
olog
y0.
116
0.15
7
Eco
logy
Eco
logy
0.90
20.
141
Bio
dive
rsity
cons
erva
tion
0.85
90.
113
0.13
5Z
oolo
gy0.
259
0.24
30.
707
Mar
ine
&fr
eshw
ater
biol
ogy
0.70
10.
288
Orn
ithol
ogy
0.67
8E
volu
tiona
rybi
olog
y0.
410
0.63
50
.145
0.25
0O
cean
ogra
phy
0.56
60.
287
0.29
70
.134
Fish
erie
s0.
525
0.11
10.
197
0.1
30Fo
rest
ry0.
517
0.37
5E
ntom
olog
y0.
194
0.36
60.
156
Che
mis
try
Che
mis
try,
mul
tidis
cipl
inar
y0.
138
0.25
10.
842
Che
mis
try,
orga
nic
0.71
5C
hem
istr
y,in
orga
nic
&nu
clea
r0.
120
0.70
10
.115
Che
mis
try,
phys
ical
0.53
00.
622
0.13
8C
hem
istr
y,ap
plie
d0.
150
0.13
20
.120
0.60
00.
141
0.48
0C
ryst
allo
grap
hy0.
338
0.60
0C
hem
istr
y,m
edic
inal
0.47
10.
164
0.51
10.
184
0.19
1Sp
ectr
osco
py0.
128
0.20
20.
496
0.1
130.
322
Eng
inee
ring
,che
mic
al0.
240
0.43
90.
268
0.27
00.
104
Che
mis
try,
anal
ytic
al0.
160
0.41
30.
158
0.12
7M
ater
ials
scie
nce,
text
iles
0.16
50.
410
0.11
10.
111
0.1
49Po
lym
ersc
ienc
e0.
289
0.37
20.
136
0.1
24M
ater
ials
scie
nce,
biom
ater
ials
0.27
20.
282
0.32
30
.156
360 JOURNAL OF THE AMERICAN SOCIETY FOR INFORMATION SCIENCE AND TECHNOLOGYFebruary 2009DOI: 10.1002/asi
-
Geo
scie
nces
Geo
scie
nces
,mul
tidis
cipl
inar
y0.
920
0.20
7G
eolo
gy0.
908
Geo
chem
istr
y&
geop
hysi
cs0.
807
Geo
grap
hy,p
hysi
cal
0.33
50.
786
0.15
6Pa
leon
tolo
gy0.
166
0.70
0M
iner
alog
y0.
137
0.64
9E
ngin
eeri
ng,g
eolo
gica
l0.
632
0.12
90.
105
Eng
inee
ring
,pet
role
um0.
104
0.20
30.
555
0.11
50.
153
Rem
ote
sens
ing
0.23
10.
471
0.1
190.
273
Met
eoro
logy
&at
mos
pher
icsc
ienc
es0.
463
0.36
5Im
agin
gsc
ienc
e&
phot
ogra
phic
0.28
00.
354
0.1
010.
285
tech
nolo
gyE
ngin
eeri
ngM
echa
nics
0.18
10.
831
0.14
2E
ngin
eeri
ng,m
echa
nica
l0.
238
0.75
70.
107
Mat
hem
atic
s,in
terd
isci
plin
ary
0.16
20.
136
0.65
20.
301
appl
icat
ions
The
rmod
ynam
ics
0.14
70.
113
0.64
90.
114
Eng
inee
ring
,mul
tidis
cipl
inar
y0.
482
0.15
20.
153
0.63
50.
107
Eng
inee
ring
,aer
ospa
ce0.
116
0.45
30.
109
Mat
eria
lssc
ienc
e,co
mpo
site
s0.
421
0.44
40
.189
Eng
inee
ring
,mar
ine
0.1
040.
387
0.26
20
.119
Con
stru
ctio
n&
build
ing
tech
nolo
gy0.
245
0.33
90.
215
0.1
56E
ngin
eeri
ng,m
anuf
actu
ring
0.24
80.
292
0.32
30
.140
0.2
50A
cous
tics
0.11
90
.106
0.29
70
.128
Mat
hem
atic
s0
.133
0.11
70.
127
0.1
250.
102
Infe
ctio
usD
isea
ses
Infe
ctio
usdi
seas
es0.
126
0.81
10.
158
Imm
unol
ogy
0.31
50.
237
0.74
9M
icro
biol
ogy
0.35
60.
652
0.21
6A
llerg
y0.
220
0.58
2V
irol
ogy
0.37
10.
573
Tro
pica
lmed
icin
e0.
565
0.36
6Pa
rasi
tolo
gy0.
300
0.11
10.
560
Myc
olog
y0.
419
0.12
70.
461
0.43
2V
eter
inar
ysc
ienc
es0.
151
0.1
030.
429
Env
iron
men
tal
Eng
inee
ring
,env
iron
men
tal
0.13
60.
176
0.11
40.
762
Scie
nces
Env
iron
men
tals
cien
ces
0.34
00.
113
0.17
90.
753
0.12
8W
ater
reso
urce
s0.
133
0.34
50.
122
0.73
0E
ngin
eeri
ng,c
ivil
0.20
50.
398
0.68
2L
imno
logy
0.54
10.
251
0.61
0A
gric
ultu
rale
ngin
eeri
ng0.
109
0.56
20.
395
Eng
inee
ring
,oce
an0
.112
0.29
80.
365
0.38
30
.115
(Con
tinu
ed)
JOURNAL OF THE AMERICAN SOCIETY FOR INFORMATION SCIENCE AND TECHNOLOGYFebruary 2009 361DOI: 10.1002/asi
-
Ap
pen
dix
.(C
onti
nued
)
Citi
ngfa
ctor
ISI
Subj
ectC
ateg
ory
Biomedicalsciences
Materialssciences
Computersciences
Clinicalsciences
Neurosciences
Ecology
Chemistry
Geosciences
Engineering
InfectiousDiseases
Environmentalsciences
Agriculture
Physics
Gen.medicine;health
Agr
icul
ture
Hor
ticul
ture
0.11
90.
101
0.80
2A
gron
omy
0.17
20.
791
Agr
icul
ture
,mul
tidis
cipl
inar
y0.
130
0.15
00.
154
0.75
7Pl
ants
cien
ces
0.29
30.
223
0.71
4Fo
odsc
ienc
e&
tech
nolo
gy0.
146
0.1
510.
177
0.1
040.
136
0.57
1So
ilsc
ienc
e0.
221
0.15
90.
263
0.45
2In
tegr
ativ
e&
com
plem
enta
rym
edic
ine
0.19
90.
138
0.32
10.
162
0.13
50.
322
0.25
8A
gric
ultu
re,d
airy
&an
imal
scie
nce
0.16
90
.122
0.1
020.
203
Mat
eria
lssc
ienc
e,pa
per
&w
ood
0.10
9
Phys
ics
Phys
ics,
mul
tidis
cipl
inar
y0.
207
0.11
40.
864
Phys
ics,
mat
hem
atic
al0.
274
0.78
1Ph
ysic
s,nu
clea
r0.
724
Phys
ics,
part
icle
s&
field
s0.
687
Phys
ics,
fluid
s&
plas
mas
0.45
60.
599
Opt
ics
0.33
70.
260
0.48
4Ph
ysic
s,at
omic
,mol
ecul
ar,&
chem
ical
0.28
20.
400
0.44
0A
stro
nom
y&
astr
ophy
sics
0.37
9N
ucle
arsc
ienc
e&
tech
nolo
gy0.
362
0.10
70.
373
Gen
eral
med
icin
e;H
ealth
care
scie
nces
&se
rvic
es0.
183
0.10
40.
786
heal
thM
edic
alet
hics
0.13
90.
721
Publ
ic,e
nvir
onm
enta
l,&
occu
patio
nal
0.11
30.
227
0.17
80.
698
heal
thM
edic
ine,
gene
ral&
inte
rnal
0.14
40.
498
0.12
90.
203
0.68
7M
edic
alin
form
atic
s0
.164
0.22
60.
100
0.1
570.
564
Nur
sing
0.46
1H
isto
ry&
philo
soph
yof
scie
nce
0.12
70.
150
0.43
4E
duca
tion,
scie
ntifi
cdi
scip
lines
0.16
80.
411
Stat
istic
s&
prob
abili
ty0.
116
0.1
840.
212
0.11
50.
150
0.19
50
.178
0.24
1
Not
e.E
xtra
ctio
nm
etho
d:Pr
inci
palc
ompo
nent
anal
ysis
.Rot
atio
nm
etho
d:V
arim
axw
ithK
aise
rN
orm
aliz
atio
n.R
otat
ion
conv
erge
din
11ite
ratio
ns.
362 JOURNAL OF THE AMERICAN SOCIETY FOR INFORMATION SCIENCE AND TECHNOLOGYFebruary 2009DOI: 10.1002/asi