180x Filetype PDF File size 0.41 MB Source: www.ncbi.nlm.nih.gov
Education and debate Whythe impact factor of journals should not be used for evaluating research Per O Seglen Institute for Studies Evaluating scientific quality is a notoriously difficult in Research and problem which has no standard solution. Ideally, pub- Higher Education lished scientific results should be scrutinised by true Summary points (NIFU), Hegdehaugsveien experts in the field and given scores for quality and • Use of journal impact factors conceals the 31, N-0352 Oslo, quantity according to established rules. In practice, difference in article citation rates (articles in the Norway most cited half of articles in a journal are cited Per O Seglen, however, what is called peer review is usually professor performed by committees with general competence 10timesasoftenastheleastcitedhalf) ratherthanwiththespecialist’sinsightthatisneededto • Journals’ impact factors are determined by BMJ 1997;314:498–502 assess primary research data. Committees tend, there- technicalities unrelated to the scientific quality fore, to resort to secondary criteria like crude of their articles publication counts, journal prestige, the reputation of • Journal impact factors depend on the authors and institutions, and estimated importance research field: high impact factors are likely in 1 journals covering large areas of basic research and relevance of the research field, making peer 23 with a rapidly expanding but short lived review as much of a lottery as of a rational process. On this background, it is hardly surprising that literature that use many references per article alternative methods for evaluating research are being • Article citation rates determine the journal sought, such as citation rates and journal impact impact factor,not vice versa factors, which seem to be quantitative and objective indicators directly related to published science. The 5 citation data are obtained from a database producedby academic positions. In the Nordic countries, journal the Institute for Scientific Information (ISI) in impact factors have, on occasion, been used in the Philadelphia, which continuously records scientific evaluation of individuals as well as of institutions and citations as represented by the reference lists of articles have been proposed, or actually used, as one of the from a large number of the world’s scientific journals. premises for allocation of university resources and 167 Thereferences are rearranged in the database to show positions. Resource allocation based on impact fac- 8 howmanytimeseachpublicationhasbeencitedwithin tors has also been reported from Canada and 9 a certain period, and by whom, and the results are Hungary and, colloquially, from several other coun- published as the Science Citation Index (SCI). On the tries. The increasing awareness of journal impact basis of the Science Citation Index and authors’ factors, and the possibility of their use in evaluation, is publication lists, the annual citation rate of papers by a already changing scientists’ publication behaviour scientific author or research group can thus be towards publishing in journals with maximum 910 calculated. Similarly, the citation rate of a scientific impact, often at the expense of specialist journals journal—known as the journal impact factor—can be that might actually be more appropriate vehicles for calculated as the mean citation rate of all the articles the research in question. 4 Given the increasing use of journal impact containedinthejournal. Journalimpactfactors,which factors are published annually in SCI Journal Citation Reports, —as well as the (less explicit) use of journal are widely regarded as a quality ranking for journals prestige—in research evaluation, a critical examination andusedextensivelybyleadingjournalsintheiradver- of this indicator seems necessary (see box). tising. Since journal impact factors are so readily Is the journal impact factor really available, it has been tempting to use them for evaluat- representative of the individual journal ing individual scientists or research groups. On the articles? assumptionthatthejournalisrepresentativeofits arti- cles, the journal impact factors of an author’s articles Relation of journal impact factor and citation rate cansimplybeaddeduptoobtainanapparentlyobjec- of article tive and quantitative measure of the author’s scientific For the journal’s impact factor to be reasonably achievement. In Italy, the use of journal impact factors representative of its articles, the citation rate of was recently advocated to remedy the purported individual articles in the journal should show a narrow subjectivity and bias in appointments to higher distribution,preferably a Gaussian distribution,around 498 BMJ VOLUME314 15FEBRUARY1997 Education and debate Problems associated with the use of 100 journal impact factors 90 • Journal impact factors are not statistically 80 representative of individual journal articles 70 • Journal impact factors correlate poorly with actual citations of individual articles % Of journal's citation rate60 • Authors use many criteria other than impact when submitting to journals 50 • Citations to “non-citable” items are erroneously included in the database 40 • Self citations are not corrected for 30 • Review articles are heavily cited and inflate the impact factor of journals 20 • Long articles collect many citations and give high journal impact factors 10 • Shortpublication lag allows many short term journal 0 self citations and gives a high journal impact factor 0102030405060708090100 • Citations in the national language of the journal are % Of journal articles (in 5% categories by annual citation rate) preferred by the journal’s authors • Selective journal self citation: articles tend to Fig 2 Cumulative contribution of articles with different citation preferentially cite other articles in the same journal rates (beginning with most cited 5%) to total journal impact. • Coverageofthedatabaseisnotcomplete Values are mean (SE) of journals in fig 1; dotted lines indicate • Booksarenotincludedinthedatabaseasasourcefor contributions of 15% and 50% most cited articles11 citations • Database has an English language bias • Database is dominated by American publications • Journal set in database may vary from year to year The uneven contribution of the various articles to • Impact factor is a function of the number of the journal impact is further illustrated in figure 2: the references per article in the research field • Research fields with literature that rapidly becomes cumulative curve shows that the most cited 15% of the obsolete are favoured articles account for 50% of the citations, and the most • Impact factor depends on dynamics (expansion or cited 50% of the articles account for 90% of the contraction) of the research field citations. In other words, the most cited half of the arti- • Small research fields tend to lack journals with high impact cles are cited, on average, 10 times as often as the least • Relations between fields (clinical v basic research, for cited half. Assigning the same score (the journal example) strongly determine the journal impact factor impact factor) to all articles masks this tremendous • Citation rate of article determines journal impact,but difference not vice versa —which is the exact opposite of what an evaluation is meant to achieve. Even the uncited articles are then given full credit for the impact of the the mean value (the journal’s impact factor). Figure 1 few highly cited articles that predominantly determine shows that this is far from being the case: three differ- the value of the journal impact factor. ent biochemical journals all showed skewed distribu- Since any large, random sample of journal articles tions of articles’ citation rates, with only a few articles will correlate well with the corresponding average of 11 12 anywherenearthepopulationmean. journal impact factors, the impact factors may seem reasonably representative after all. However, the corre- lation between journal impact and actual citation rate 600 of articles from individual scientists or research groups Biochimica Biophysica Acta 912 Biochemical Journal is often poor (fig 3). Clearly, scientific authors do not Journal of Biological Chemistry necessarily publish their most citable work in journals 500 of the highest impact, nor do their articles necessarily No of articles cited match the impact of the journals they appear in. 400 Although some authors may take journals’ impact fac- tors into consideration when submitting an article, other factors are (or at least were) equally or more 300 important,suchasthejournal’ssubjectareaanditsrel- evance to the author’s specialty, the fairness and rapid- ity of the editorial process, the probability of 200 acceptance,publication lag, and publication cost (page 13 charges). 100 Journal impact factors are representative only when the evaluated research is absolutely average (relative to the journals used), a premise which really 0 makes any evaluation superfluous. In actual practice, however, even samples as large as a nation’s scientific ≤1 2-3 4-5 6-7 8-9 30 output are far from being random and representative 10-1112-13 14-15 16-1718-1920-2122-2324-2526-2728-29 ≥ No of citations per article per year of the journals they have been published in: for exam- Fig 1 Citation rates in 1986 or 1987 of articles published in three ple, during the period 1989-93, articles on general biochemical journals in 1983 or 1984, respectively11 medicine in Turkey would have had an expected citation rate of 1.3 (relative to the world average) on the BMJ VOLUME314 15FEBRUARY1997 499 Education and debate articles. If correction were made for article length, 10 r=0.05 10 r=0.27 “communications” journals like Biochemical and Bio- physical Research Communications and FEBS Letters 5 5 wouldgetimpactfactorsashighas,orhigherthan,the high impact journals within the field, like Journal of 20 21 2 2 Biological Chemistry. Theuseofanextremelyshorttermindex(citations 1 1 to articles published only in the past two years) in cal- No of citations per article per year culating the impact factor introduces a strong 0.5 0.5 temporalbias,withseveralconsequences.Forexample, articles in journals with short publication lags will con- 0.2 0.2 tain relatively many up to date citations and thus con- tribute heavily to the impact factors of all cited 0 0 journals. Since articles in a given journal tend to cite 22 r=0.44 r=0.63 articles from the same journal, rapid publication is 10 50 self serving with respect to journal impact, and signifi- 23 5 20 cantly correlated with it. Dynamicresearchfields with high activity and short publication lags, such as 10 biochemistry and molecular biology, have a corre- 2 5 spondingly high proportion of citations to recent publications —andhencehigher journal impact factors 1 2 23 24 No of citations per article per year —than, for example, ecology and mathematics. 1 Russian journals, which are cited mainly by other 0.5 25 0.5 Russian journals, are reported to have particularly long publication lags, resulting in generally low impact 0.2 26 0.2 factors. Pure technicalities can therefore account for 0 0 several-fold differences in journal impact. 0.5 1 2 5 10 0.5 1 2 5 10 20 Limitations of the database Impact factor of journal Impact factor of journal Fig 3 Correlation between article citation rate and journal impact for four authors12 The Science Citation Index database covers about 8 3200 journals ; the estimated world total is about 27 126000. The coverage varies considerably between basis of journal impact, but the actual citation was only research fields: in one university, 90% of the chemistry 14 faculty’s publications, but only 30% of the biology fac- 0.3. The use of journal impact factors can therefore 28 beasmisleadingforcountries as for individuals. ulty’s publications, were in the database. Since the impactfactorofanyjournalwillbeproportionaltothe Journal impact factors are calculated in a way that database coverage of its research field, such discrepan- causes bias cies mean that journals from an underrepresented Apart from being non-representative, the journal field that are included will receive low impact factors. impact factor is encumbered with several shortcom- Furthermore, the journal set in the database is not ings of a technical and more fundamental nature. The constant but may vary in composition from year to factor is generally defined as the recorded number of year.24 29 In many research fields a substantial fraction citations within a certain year (for example, 1996) to of scientific output is published in the form of books, the items published in the journal during the two pre- whicharenotincludedassourceitemsinthedatabase; ceding years (1995 and 1994), divided by the number 30 they therefore have no impact factor. In mathematics, of such items (this would be the equivalent of the aver- leading publications that were not included in the Sci- age citation rate of an item during the first and second ence Citation Index database were cited more calendar year after the year of publication). However, frequently than the leading publications that were the Science Citation Index database includes only nor- 31 included. Clearly, such systematic omissions from the mal articles, notes, and reviews in the denominator as databasecancauseseriousbiasinevaluationsbasedon citable items, but records citations to all types of docu- impact factor. ments (editorials, letters, meeting abstracts, etc) in the Thepreference of the Science Citation Index data- numerator; citations to translated journal versions are 28 baseforEnglishlanguagejournals willcontributetoa 15-17 even listed twice. Because of this flawed computa- lowimpactfactorforthefewnon-Englishjournalsthat 32 tion, a journal that includes meeting reports, are included, since most citations to papers in interesting editorials, and a lively correspondence sec- languages other than English are given by other tion can have its impact factor greatly inflated relative 25 27 33 papers in the same language. The Institute for to journals that lack such items. Editors who want to Scientific Information’s database for the social sciences raise the impact of their journals should make frequent contained only two German social science journals, reference to their previous editorials, since the 34 whereas a German database contained 542. Specifi- database makes no correction for self citations. The cally, American scientists, who seem particularly prone inclusion of review articles, which generally receive 33 35 to citing each other, dominate these databases to 17 18 many more citations than ordinary articles, is also such an extent (over half of the citations) as to raise recommended. Furthermore, because citation rate is both the citation rate and the mean journal impact of 19 14 roughly proportional to the length of the article, American science 30% above the world average, the journals might wish to publish long, rather than short, rest of the world then falling below average.This bias is 500 BMJ VOLUME314 15FEBRUARY1997 Education and debate aggravated by the use of a short term index: for exam- Table 1 Journal impact factors and research field ple, in American publications within clinical medicine, 83% of references in the same year were to other Journal 1986 1987 papersbyAmericanscientists(manyofthemundoubt- Annual Review of Biochemistry 31.6 35.1 edly self citations), a value 25% higher than the stable Annual Review of Immunology 26.5 25.2 level reached after three years (which would, inciden- Annual Review of Cell Biology 14.1 22.8 tally, also be biased by self citations and citations of Annual Review of Genetics 14.0 14.3 33 Annual Review of Neuroscience 15.4 13.7 other American work). Thus,both the apparent qual- Annual Review of Pharmacology 10.1 9.9 ity lead of American science and the values of the vari- Annual Review of Physiology 7.8 9.1 ous journal impact factors are, to an important extent, Annual Review of Biophysics 7.2 7.7 determined by the large volume, the self citations, and Annual Review of Microbiology 4.9 6.4 27 the national citation bias of American science, in combination with the short term index used by the Science Citation Index for calculating journal impact heavily on basic science,but not vice versa.The result is factors. that basic medicine is cited three to five times more than clinical medicine, and this is reflected in journal 42 44 45 Journal impact factors depend on the research field impact factors. The outcome of an evaluation Citation habits and citation dynamics can be so differ- based on impact factors in medicine will therefore ent in different research fields as to make evaluative depend on the position of research groups or 33 comparisons on the basis of citation rate or journal institutions along the basic-clinical axis. impact difficult or impossible. For example,biochemis- In measures of citation rates of articles, attempts to try and molecular biology articles were cited about five take research field into account often consist of 33 expressingcitationraterelativetosomecitationimpact times as often as pharmacy articles. Several factors 46 have been found to contribute to such differences specific to the field. Suchfield corrections range from amongfieldsofresearch. simply dividing the article’s citation rate by the impact 28 The citation impact of a research field is directly factor of its journal (which punishes publication in proportional to the mean number of references per high impact journals) to the use of complex, author 47 48 article,whichvariesconsiderablyfromfieldtofield(itis specific, field indicators based on reference lists twice as high in biochemistry as in mathematics, for (which punishes citations to high impact journals). 24 However,field corrections cannot readily be applied to example). Within the arts and humanities, references to articles are hardly used at all, leaving these research journal impact factors, since many research fields are 36 dominatedbyoneorafewjournals,inwhichcasecor- fields (and others) virtually uncited, a matter of considerable consternation among science administra- rections might merely generate relative impact factors 37 of unit value. Even within large fields, the tendency of tors unfamiliar with citation kinetics. Inhighlydynamicresearchfields,suchasbiochem- journals to subspecialise with certain subjects is likely istry and molecular biology, where published reports to generate significant differences in journal impact: in rapidly become obsolete, a large proportion of a single biochemical journal there was a 10-fold differ- 19 citations are captured by the short term index used to enceincitation rates in subfields. calculate journal impact factors, as previously 38 discussed —but fields with a more durable literature, Is the impact of an article increased by such as mathematics, have a smaller fraction of short publication in a high impact journal? term citations and hence lower journal impact factors. This field property combines with the low number of It is widely assumed that publication in a high impact references per article to give mathematics a recorded journal will enhance the impact of an article (the “free citation impact that is only a quarter that of ride” hypothesis). In a comparison of two groups of 24 biochemistry. scientific authors with similar journal preference who Inyoungandrapidlyexpandingresearchfields,the differed twofold in mean citation rate for articles, how- number of publications making citations is large ever, the relative difference was the same (twofold) relative to the amount of citable material, leading to throughout a range of journals with impact factors of 12 high citation rates for articles and high journal impact 0.5 to 8.0. If the high impact journals had contributed 39 40 factors for the field. “free” citations, independently of the article contents, In a largely self contained research field, the mean the relative difference would have been expected to 49 article (or journal) citation rate is independent of the diminish as a function of increasing journal impact. 41 size of the field, buttheabsoluterangewillbewiderin These data suggest that the journals do not offer any a large field, meaning higher impact factors for the top free ride. The citation rates of the articles determine 42 journals. Such differences become obvious when the journal impact factor (a truism illustrated by the comparing review journals, which tend to top their good correlation between aggregate citation rates of field (table 1). Leading scientists in a small field may article and aggregate journal impact found in these thus be at a disadvantage compared with their data), but not vice versa. colleagues in larger fields, since they lack access to If scientific authors are not detectably rewarded 43 with a higher impact by publishing in high impact journals of equally high citation impact. Most research fields are, however, not completely journals, why are we so adamant on doing it? The self contained,themostimportantfieldfactorprobably answer,ofcourse,isthataslongastherearepeopleout being the ability of a research field to be cited by adja- there who judge our science by its wrapping rather cent fields. The relation between basic and clinical than by its contents, we cannot afford to take any medicine is a case in point: clinical medicine draws chances. Although journal impact factors are rarely BMJ VOLUME314 15FEBRUARY1997 501
no reviews yet
Please Login to review.