Revisiting the journal impact factor
←
→
Page content transcription
If your browser does not render page correctly, please read the page content below
Revisiting the journal impact factor Vera Morgan and Aleksandar Janca INTRODUCTION I n 1955, Eugene Garfield proposed “a bibliographic system for science literature that can eliminate the uncritical citation of fraudulent, incomplete, or obsolete data by making it possible for the conscien- tious scholar to be aware of criticisms of earlier papers”.1 This led to the publication of the Science Citation Index (SCI) by the Institute for Scien- tific Information (ISI) and the development, by Garfield and Sher of the ISI, of the journal impact factor to facilitate the selection of journals for the SCI and Current Contents2 (Table 1). Some 44 years later, and 36 years after the publication of the first Science Citation Index, Garfield wrote that “[l]ike nuclear energy, the impact factor has become a mixed blessing”.2 Hecht et al are more unequivocal in their criticism, describ- ing the impact factor as “a misnamed, misleading, misused measure”.3 This paper revisits the concept of the impact factor. It identifies two main uses and several limitations in its use. It argues that distinguish- ing the two uses of the impact factor helps one to decide whether the journal impact factor is valuable or whether it is an ill-conceived meas- ure that should be abandoned. WHAT IS A JOURNAL’S IMPACT FACTOR The impact factor of a journal is an ISI measure found in the Journal Citation Reports (JCR), an index of published journals in the sciences and social sciences produced annually by the ISI. It is one of several values in the JCR based on the frequency with which articles published by a journal are cited. The impact factor, the most common of these values, is used to rank journals both overall and within a discipline area (Table 1). The impact factor measures the average number of times articles pub- lished within the previous two years by a journal are cited in the current Australasian Psychiatry • Vol 8, No 3 • September 2000 Table 1: Terms and abbreviations Vera Morgan, BA, DipEd, MSocSc ISI Institute for Scientific Information which publishes: Research Fellow University of Western Australia SCI Science Citation Index Department of Psychiatry and Behavioural Science Perth, Western Australia Aleksandar Janca, MD, MSc, FRCPsych, FRANZCP JCR Journal Citation Report Associate Professor of Psychiatry University of Western Australia Impact factor The impact factor measures the average number of times articles Department of Psychiatry and Behavioural Science published within the previous two years by a journal are cited in Perth, Western Australia Correspondence: Vera Morgan, University of Western the current year. Australia, Department of Psychiatry and Behavioural Science, Medical Research Foundation Building, Rear 50 Murray Street, Cited half-life The cited half-life measures the number of publication years Perth, WA 6000. Tel: 08 9224 0235 preceding the current year which account for 50% of total citations Fax: 08 9224 0285 within a journal in the current year. Email: vmorgan@cyllene.uwa.edu.au 230
year covered by the JCR. For example, the 1998 However, the JCR also adds that its citation scores impact factor for an individual journal is the number “should not be relied on as the sole source of infor- of citations in 1998 of articles published by the jour- mation when comparing and evaluating publica- nal in 1996–97 divided by the number of articles pub- tions. The quantitative citation data it reports are lished by the journal in the same two-year period. intended to complement, not replace, traditional Only original research articles, review articles and qualitative and subjective inputs, such as peer sur- technical notes are included in JCR counts of articles veys and specialist opinions”.5 Nonetheless, there is in the denominator. However, counts of citations in growing evidence that the impact factor is used in the numerator also include editorials, letters, academic appointments and promotion,6,7 and in abstracts and the like. resource allocation.7 It would appear that some scientific committees, while recognising that a sim- A second JCR measure is the cited half-life. The cited ple count of articles published is poor evidence of half-life measures the number of publication years the quality of research output, are coming to rely on preceding the current year covered by the JCR which journal impact factors as a shortcut to assessing the account for 50% of total citations within a journal in quality of individual papers and their authors. Hecht the current year. Only journals with at least 100 cita- argues that it has had a particularly strong impact in tions are listed. For example, in 1998 a journal may Western Europe, possibly in response to a need for have included citations going as far back as 1964 but an objective system of review following a history of 50% of the citations in the journal were for items “favouritism, nepotism, and social-rank privileges”.3 published between 1993 and 1998. The JCR formula Benítez-Bribiesca observes that its popularity is calculates a cited half-life of some 7 years, with the growing in developing countries who are endeavour- exact figure based on the cumulative percentage of ing to ensure that their own research committees citations per year. apply the same rigorous standards found in highly developed countries.6 The JCR complements the SCI. The SCI compiles a searchable list of all substantive items published in journals during the period covered by the index LIMITATIONS IN THE USE OF IMPACT (Source Index), and further compiles all citations to FACTORS that item (Citation Index). These indexes are organ- Limitations due the ISI method of calculation ised in two parts: by items cited and by articles cit- ing those items, both sorted on first author. Because There are a number of inherent limitations in the ISI calculation of the impact factor which call into ques- there is also a subject index (Permuterm Subject tion its usefulness. First, the source database, the Index), it is possible to trace published articles SCI, is not comprehensive. It does not include books through the subject matter. Thus the SCI allows and, in 1997, it only covered some 3200 journals out researchers to determine whether their publication of an estimated world total of 126,000.7 There are has been cited by others, and to trace those refer- strong biases as to which journals are covered. There ences. It is also a means of following backwards or is a preponderance of journals written in English forwards work published by a particular author or in and, if in English, published in North America.7 a particular field of research. Indexes, similar to the Opthof cites de Jong and Schaper’s 1996 data for science index, have been published for the arts and clinical cardiovascular papers.8 They analysed cita- humanities, and for the social sciences. The JCR is tions for 137,019 papers published between 1981 and compiled using data from these three indices. 1992 from the G7, seven major industrial democra- cies, Canada, France, Germany, Italy, Japan, United THE USE OF THE IMPACT FACTOR Australasian Psychiatry • Vol 8, No 3 • September 2000 Kingdom and USA, as well as 7 smaller European Garfield argues that the primary aim of the impact countries. Some 46% of these papers were never factor is to help libraries decide what to purchase cited, ranging from 31% for Norway to 69% for and authors decide where to submit their manu- Japan. The average number of citations per paper scripts.2 The JCR also recommends its general use for was 7.5 for USA down to 2.0 for Japan. Moreover, it tracking bibliographic trends, as well as a source of appears that discipline area also affects inclusion on evaluative data for the editorial and publication the ISI database. One institution found that, in 1987, boards of journals. Editors of journals such as the the database included 90% of publications from its British Journal of Psychiatry endorse the use of the faculty of chemistry but only 30% of those from the journal impact factor as a measure of journal status: faculty of biology.7 “Readers and future authors may like to know that Second, articles that are counted in the denominator work published in the Journal is cited relatively of the impact factor equation — substantive citable frequently — an indication that articles in the Jour- articles — may be hard to classify as such. This has nal are read and have an impact on the psychiatric led to incorrect classifications which, in turn, affect community”.4 the impact factor which decreases in value as the 231
denominator increases, assuming the numerator Other aspects of journals will affect the impact fac- remains static. For example, Joseph and Hoey calcu- tor. Elite generalist journals can be more selective lated there were between 176 and 208 citable articles about what they publish as they are not limited to a in the Canadian Medical Association Journal (CMAJ) specific field, nor do they have obligations to pub- for the period 1995–96 compared to the JCR’s count lish the work of small and/or foreign research of 350. The erroneous counting, by the ISI, of certain groups.3 Journals published in English and, if in Eng- items as citable was a mistake identified by the CMAJ lish, in North America are not only more likely to be which will be corrected in the JCR from 2000.9 included on the ISI database for the calculation of Third, citations which are the basis of the numerator the impact factor (see above), they are also more may be positive or negative, a consideration not likely to be read and cited for linguistic and cultural included in the calculation of the impact factor reasons.4,7,8 which merely counts the number of citations. Finally, editorial policy may be influential. For Finally, a journal’s impact factor will be adversely example, the time taken to review manuscripts is affected either if it has a change of name over the important given a two-year limit on the calculation two-year calculation period as the two journal titles of citations.12 Self-citations will also boost a journal are not linked for the purposes of calculating the impact factor;7 it has been demonstrated that two impact factor,10 or if names in a citation are mis- spurious self-citations per source article will increase spelled, a not insignificant problem which affected a journal impact factor of 0.1 to 2.1.13 10% of citations in cardiovascular research in 1994.8 Biases arising out of journal characteristics QUALITY AND THE IMPACT FACTOR Other problems in the use of the impact factor arise Given these limitations, what can be said of the use out of biases affecting the algorithm for calculating of the impact factor as a measure of quality. Seglen the impact factor introduced through journal char- argues, with good evidence, that the impact factor is acteristics. Many of these biases arise because the useful in assessing the quality of a journal but not the source for citations in the impact factor numerator is quality of any individual paper or author. For one more broadly defined than the source for substantive thing, he contends, the impact factor cannot items in the denominator; journal characteristics represent author impact where only 15% of journal that increase the numerator at a greater rate than the articles account for 50% of citations.7 However, the denominator will tend to result in higher impact analysis of simultaneous publication in multiple factors. journals suggests that, while the level of citation is more influenced by the quality of an article than by The nature of the research published in a journal will the journal impact factor, journal visibility may exert a strong influence on its impact factor. increase the citation rate by up to 80%.8 This, in Research in the basic sciences is more likely to be cited compared to articles with clinical and indus- turn, makes these prestigious journals highly attrac- trial applications, or relating to health care pro- tive to good researchers, be it for the prestige or to grams, which may be widely read and applied but ensure wider dissemination of their findings or not cited much.6,7,10 Likewise, published research in because publication in high impact journals will disciplines in which knowledge is expanding rapidly overtly or covertly increase their chances of gaining and publication is a high priority such as genetics funding or promotion. “However, if the papers we will be cited with greater frequency within the two- submit are not of high quality, they will not be year time limit for the calculation of the impact accepted by prestigious journals and we have to sub- Australasian Psychiatry • Vol 8, No 3 • September 2000 factor than publications in other disciplines such as mit articles of moderate quality to journals of inferior mathematics.6,7,10 The cited half-life of journals gives impact”, writes Hoeffel.14 a good indication of the heterogeneity that exists These trends lie at the heart of warnings that high between disciplines in the timeframe within which impact factor journals will attract the highest quality articles will be cited; in some cases, that heterogene- contributions to the detriment of other journals, and ity can even be seen within a discipline, for example, their use as a measure of quality in scientific evalua- oral science.10 tions will have important repercussions for new The type of items published in a journal can also fields, young researchers and research groups in lead to biases. Review articles generate more cita- developing countries.6 They also underlie a concern tions than research papers, because it is easier to ref- that, although high impact journals are not always erence a review article than many separate original the most appropriate vehicle for some publica- articles.3,7 Similarly published abstracts are often tions,6,15 an excellent article in an average impact cited but do not affect the denominator in the cal- journal may be poorly received compared to an aver- culation of the journal impact factor.11 age article in a high impact journal.3 232
EXPLOITING THE IMPACT FACTOR: CONCLUSION EFFECT ON EDITORIAL POLICY Two primary uses of a journal impact factor have A further danger is that a journal may introduce been identified. The first use is in assessing the changes merely to boost its impact factor. Hecht impact of a journal, either overall or within a field. et al. warn that “[e]ditorial policies once determined This use seems inevitable and even warranted in a by scientists-editors may increasingly be dictated by period where the number of journals is expanding executives and accountants”. It is inevitable that rapidly and time for evaluation is limited. However, some provisos are essential. Given the limitations of journals should begin to use impact factors as mar- the database and the effect of journal characteristics keting tools, and evidence of this use is growing.3,4,16 on the impact factor, it is important for users to The number of journals has increased exponentially become familiar with the mechanisms underlying over the last half-century. At the same time, the calculation of the impact factor, in order to bet- researchers, faced with fewer resources and substan- ter understand its virtues and deficits. In particular, tially larger administrative roles, are increasingly it is essential to recognise that, while it appears to judicious in their use of time and money. Similar hold that journals with high impact factors are of constraints apply to departmental and institutional high quality, the converse is not necessarily true and libraries. In these circumstances, journals will do that at least some high quality journals will have what they can to win their share of readers. There is moderate, low or missing impact factors for any of some evidence that this includes, not just marketing the reasons already outlined. by a journal of its impact factor, but proactive changes to a journal format to increase its impact At the same time, changes to journal policy need to factor. The Canadian Medical Association Journal be monitored to ensure that the quest for a higher notes the importance of increasing its impact factor impact factor is not at the expense (directly or indi- rectly) of substantive changes to the quality of scien- to attract authors.9 One measure planned is a shorter tific publication. time between receipt of a manuscript and publica- tion.15 The Lancet now uses explicit citations to orig- Whether one should go so far as to separate review inal articles in its letters of comment rather than journals from non-review journals and elite journals including them in the text as it did formerly, thus from discipline-specific journals, and rename the increasing the number of self-citations.13 Other jour- impact factor a “citation rate index” in keeping with nals such as the FASEB Journal (the journal of the its actual role3 is open to debate. However, one sim- Federation of American Societies for Experimental ple procedure available to all users is to consider the Biology) and Nature have reclassified some items so discipline area when comparing impact factors that they are not counted as source items for the cal- across journals. Standardising the impact factor culation of the denominator but may be used in the across discipline areas takes into account at least count of citations. Thus, for example, when meeting some of the influences affecting the citation rate abstracts changed from being counted as source to within a particular area. The simplest way of doing being counted as non-source in the FASEB Journal, this is to divide the impact factor of any journal by its impact factor rose from 0.24 in 1988 to 18.3 in the impact factor of the journal ranked top within the same discipline area. A multiplier can be used to 1989. 13 Numerous writers have pointed to an set the maximum possible range. increase in the publication of review articles in order to boost the number of citations; Hecht et al. observe The second use of the impact factor is to evaluate that 15 of the top ranked journals do not publish individuals and their work on the basis of the impact Australasian Psychiatry • Vol 8, No 3 • September 2000 fresh results, only compilations of reviews and sum- factor of journals in which their publications appear. maries of past research.3 This extension of what amounts to an ecological fal- lacy into the area of citation analysis is highly con- Advances in computer technology may lead to new troversial. It accepts that the quality of published ways of enhancing one’s impact factor. For example, research can be evaluated from its “wrapping” rather a journal that alerts its readership to its current con- than from an examination of its “contents”.7 The tents and provides easy and timely access to full text use of the impact factor in these circumstances may articles may be indirectly increasing the number of be unwarranted where other options are available. citations made to it. Unfortunately, while the published literature has The trend to editorial changes to boost one’s impact debated extensively the use of the impact factor as a factor is relatively new. Many changes documented means of predicting author quality, relatively little have been cosmetic rather than substantive. discussion has focussed on the development of alter- Whether such changes ultimately improve or detract native means of researcher evaluation. One such from the substantive quality of a journal remains to alternative is to further develop the use of actual be seen. author citation data in the SCI. The use of author 233
Table 2: JCR classification of journals within psychiatry and their rankings in 1998. 1. Arch Gen Psychiatry 9.398 40. Int J Eat Disorder 1.139 2. Am J Psychiatry 5.939 41. J Psychosom Res 1.129 3. J Clin Psychopharmacol 5.338 42. Prog Neuro-Psychoph 1.114 4. Mol Psychiatry 4.756 43. Can J Psychiatry 1.058 5. Schizophr Bull 4.455 44. J Psychiatry Neurosci 0.966 6. Neuropsychopharmacology 4.318 45. Eur Arch Psychiatry Clin Neurosci 0.959 7. Dementia 4.148 46. Aust NZ J Psychiatry 0.847 8. J Clin Psychiatry 4.073 47. Neuropsychobiology 0.846 9. J Am Acad Child Adolesc Psychiatry 3.732 48. J Intellect Disabil Res 0.754 10. Br J Psychiatry 3.503 49. Dement Geriatr Cogn 0.710 11. Psychol Med 3.124 50. Br J Med Psychol 0.702 12. Psychosom Med 3.046 51. Nervenarzt 0.696 13. Psychopharmacology 3.032 52. Behav Med 0.649 14. J Neurol Neurosurg Psychiatry 2.938 53. Stress Medicine 0.648 15. Psychopharmacol Bull 2.590 54. Int J Psychiatry Med 0.630 16. Schizophr Res 2.496 55. J Psychosom Obst Gyn 0.606 17. Biol Psychiatry 2.405 56. Hum Psychopharm Clin 0.584 18. Pharmacopsychiatry 2.304 57. Convulsive Ther 0.500 19. Am J Orthopsychiatry 2.267 58. Neuropsy Neuropsy Be 0.500 20. Psychother Psychosom 2.103 59. Verhaltenstherapie 0.486 21. J Neuropsychiatry Clin Neurosci 1.697 60. Psychopathology 0.474 22. Addiction 1.620 61. Subst Use Misuse 0.457 23. J Child Adol Psychop 1.609 62. Child Psychiatry Hum De 0.452 24. J Affect Disord 1.586 63. Fortschr Neurol Psychiatr 0.441 25. Psychiatr Serv 1.570 64. Eur Psychiat 0.394 26. Int Clin Psychopharmacol 1.557 65. Psychiat Clin Neuros 0.382 27. Acta Psychiatr Scand 1.554 66. J Geriatr Psychiatry 0.375 28. Psychosomatics 1.541 67. Encephale 0.290 29. Exp Clin Psychopharm 1.510 68. Neurol Psychiat Br 0.216 Australasian Psychiatry • Vol 8, No 3 • September 2000 30. Drug Alcohol Depen 1.485 69. Nervenheilkunde 0.211 31. Psychiatry Res 1.424 70. Ann Med Psychol 0.208 32. J Psychiat Res 1.362 71. Arq Neuropsiquiatr 0.141 33. Psychiat Res-Neuroim 1.329 72. Neuropsychiatrie 0.125 34. Psychiatry 1.255 73. Acta Neuropsychiatr 0.089 35. J Nerv Ment Dis 1.250 74. Zh Nevropatol Psikhiatr 0.080 36. Compr Psychiatry 1.234 75. Giorn Neuropsi Evol 0.031 37. J Geriatr Psychiatry Neurol 1.222 76. Actas Luso Esp Neurol Psiquiatr Cienc Afines 0.013 38. Z Psychosom Med Psycoanal 1.209 77. J Ect — 39. Gen Hosp Psychiatry 1.144 234
citation data is not without its problems, many of 3. Hecht F, Hecht B, Sandberg A. The journal ‘impact factor’: a misnamed, mislead- which relate to limitations inherent in the SCI data- ing misused measure. Cancer Genetics and Cytogenetics 1998; 104: 77–81. base. One author may appear under multiple names, 4. Howard L, Wilkinson G. Impact factors of psychiatric journals. British Journal of particularly authors with particle or compound Psychiatry 1997; 170: 109–112. names, those with names that are not English, or 5. Institute/for/Scientific/Information. 1998 JCR Journal Citation Reports. Science those who have changed their name. A further prob- Edition. Philadelphia: Institute for Scientific Information; 1999. lem occurs where an author is cited incorrectly — or 6. Benítez-Bribiesca L. The impact factor of medical journals: its use and misuse. even with variant initials. Other shortcomings aris- Archives of Medical Research 1999; 30: 161–162. ing out of the incomplete coverage of the ISI data- 7. Seglen P. Why the impact factors of journals should not be used for evaluating base and raised in relation to the impact factor also research. British Medical Journal 1997; 314: 497. apply here. However, there is some evidence that the 8. Opthof T. Sense and nonsense about the impact factor. Cardiovascular Research number of citations correlates reasonably well, if not 1997; 33: 1–7. perfectly, with peer esteem.8 Furthermore, it may be possible to shift from mere numeration to the devel- 9. Joseph K, Hoey J. CMAJ’s impact factor: room for recalculation. Canadian Med- ical Association Journal 1999; 161: 977–978. opment of more detailed longitudinal profiles of researcher performance. Researchers subject to scien- 10. Linde A. On the pitfalls of journal ranking by impact factor. European Journal of tific scrutiny have an obligation to develop and Oral Sciences 1998; 106: 525–526. refine the use of suitable performance indicators for 11. van Zanten S. Impact of abstracts and short reports. Letter. Canadian Medical their own industry. And, once established, they have Association Journal 2000; 162: 489. a right to demand that these indicators be used as 12. Hansson S. Impact factor as a misleading tool in the evaluation of medical jour- the basis for a more equitable evaluation of scientific nals. Letter. Lancet 1995; 346: 906. merit than currently available through the use of the 13. Gowrishankar J, Divakar P. Sprucing up one’s impact factor. Letter. Nature 1999; impact factor. 401: 321–322. REFERENCES 14. Hoeffel C. Journal impact factors. Allergy 1998; 53: 1225. 1. Garfield E. Citation indexes to science: a new dimension in documentation through 15. Weir E, Joseph K. How quickly does CMAJ evaluate submissions? Canadian Med- association of ideas. Science 1955; 122: 108–111. ical Association Journal 1999; 161: 985. 2. Garfield E. Journal impact factor: a brief overview. Canadian Medical Association 16. Howard L, Wilkinson G. Impact factors of psychiatric journals. British Journal of Journal 1999; 161: 979–980. Psychiatry 1998; 172: 457. Australasian Psychiatry • Vol 8, No 3 • September 2000 235
Copyright of Australasian Psychiatry is the property of Sage Publications, Ltd. and its content may not be copied or emailed to multiple sites or posted to a listserv without the copyright holder's express written permission. However, users may print, download, or email articles for individual use.
You can also read