Content uploaded by Birger Hjørland
Author content
All content in this area was uploaded by Birger Hjørland
Content may be subject to copyright.
Practical potentials of Bradford’s
law: a critical examination of the
received view
Jeppe Nicolaisen and Birger Hjørland
Royal School of Library and Information Science, Copenhagen, Denmark
Abstract
Purpose – The purpose of this research is to examine the practical potentials of Bradford’s law in
relation to core-journal identification.
Design/methodology/approach – Literature studies and empirical tests (Bradford analyses).
Findings – Literature studies reveal that the concept of “subject” has never been explicitly addressed
in relation to Bradford’s law. The results of two empirical tests (Bradford analyses) demonstrate that
different operationalizations of the concept of “subject” produce quite different lists of core-journals.
Further, an empirical test reveals that Bradford analyses function discriminatorily against minority
views.
Practical implications – Bradford analysis can no longer be regarded as an objective and neutral
method. The received view on Bradford’s law needs to be revised.
Originality/value – The paper questions one of the old dogmas of the field.
Keywords Identification, Information science, Serials, Bibliographies
Paper type Research paper
1. Introduction: Bradford’s law
Bradford’s law of scattering is often mentioned together with Zipf’s law and Lotka’s
law as one among the three most important bibliometric laws, and is often considered
the best model or example of scientific research that is available within library and
information science (LIS). Bradford’s law states that documents on a given “subject” is
distributed (or scattered) according to a certain mathematical function so that a certain
growth in papers on a subject requires a certain growth in the number of journals. The
numbers of the groups of journals to produce nearly equal numbers of articles is
roughly in proportion to 1: n: n
2
..., where n is called the Bradford multiplier[1]. In
other words, Bradford’s law states that a small core of, for example, journals have as
many papers on a given subject as a much larger number of journals, n, which again
has as many papers on the subject as n
2
journals.
Bradford himself provided both a graphical and a verbal formulation of his law that
have later been found not to be mathematical equivalent. The exact mathematical
function has been subject to much subsequent research, and the very question what a
Bradford distribution is has been debated.
The current issue and full text archive of this journal is available at
www.emeraldinsight.com/0022-0418.htm
The authors gratefully acknowledge the financial support from the Danish Ministry of Culture
(grant no. A2004 06-026) and a travel grant from the Nordic Research School in Library and
Information Science (NORSLIS) used when the authors presented their preliminary ideas for this
paper at the CoLIS 5 conference in Glasgow.
Practical
potentials of
Bradford’s law
359
Received 12 January 2006
Revised 22 March 2006
Accepted 22 March 2006
Journal of Documentation
Vol. 63 No. 3, 2007
pp. 359-377
q Emerald Group Publishing Limited
0022-0418
DOI 10.1108/00220410710743298
Bradford’s law has been used as an argument about how to build collections, how to
select journals to be indexed in bibliographies, how to measure the coverage of
bibliographies, how to solve practical problems related to information seeking and
retrieval, and by Bradford himself as an argument for a new way to organize
bibliographical work and scientific documentation. According to the received view on
Bradford’s law[2], this law may consequently help solving many of the practical
problems facing the practitioners of our profession. The basic assumption of the
advocates of the received view is that Bradford’s law functions as a completely neutral
and objective method.
We see a number of problems associated with the received view. In this paper we
will concentrate on two of these. Our first objection against the received view is that the
way one chooses to operationalize the concept of subject, when conducting Bradford
analyses, will influence on the results of the very same. Consequently, Bradford’s law
does not automatically function as a neutral method. On the contrary, the results of
utilizing Bradford analysis as a method for identifying the core information sources of
any subject, field, or discipline will depend on the way “subject” is operationalized. We
will illustrate this to be true by two examples. Our second objection against the
received view is that the selection of information sources based on
Bradford-distributions tends to favorite dominant theories and views while
suppressing views other than the mainstream at a given time. Thus, Bradford’s law
does not function as an objective method either. We will present empirical evidence
that illustrate this to be true. However, before going into details with these two
objections, we will first provide a very short review of the received view on Bradford’s
law.
2. The received view
According to the received view there are a number of practical potentials of Bradford’s
law. B.C. Brookes was among the first to address these potentials. In a short note in
Nature he wrote that the law “seems to offer the only means discernible at present to
reducing the present quantitative untidiness of scientific documentation, information
systems and library services to a more orderly state of affairs capable of being
rationally and economically planned and organized” (Brookes, 1969, p. 953).
Several commentators have suggested using Bradford’s law to solve practical
journal collection management problems. The basic idea is to conduct Bradford
analyses of journals – i.e. to sort the journals in Bradford zones – and then identify
which belong to the core and which does not. Any Bradford analysis involves three
steps (Diodato, 1994, pp. 16-17):
(1) identify many or all items (usually articles) published in this field;
(2) list the sources (usually journals) that publish the articles (or items) in rank
order beginning with the source that produces the most items; and
(3) while retaining the order of the sources, divide this list into groups (or zones) so
that the number of items produced by each group of sources is about the same.
Nisonger (1998, pp. 139-40) argues in his textbook Management of Serials in Libraries
that the following points are some of the “most obvious potentials” of Bradford
analyses:
JDOC
63,3
360
.
selection/deselection;
.
defining the core;
.
collection evaluation;
.
the law of diminishing returns;
.
calculation of cost at various coverage; and
.
setting priorities among journals.
Other commentators have suggested using Bradford’s law to solve practical problems
related to information seeking and retrieval. Howard D. White (1981) proposed an
automatic option for sorting the output from online searches of journal literature,
which he argued would help online users. What he had in mind was a “computerized
sorting of hits by the journals in which they appear, and then of journals, high to low,
by the number of hits appearing in each” (White, 1981, p. 47). He termed the procedure
“Bradfordizing”, and argued that “the ability to retrieve items selectively by journal
after learning contributing titles and their yields, would seem to be the greatest single
advantage of the proposed option” (White, 1981, p. 50). The reason for his optimism is
spelled out in the article. According to White (1981) it is easy to imagine situations in
which the searcher would want to retrieve hits only in the core journals of a literature.
He mentions that it is often troublesome to track down the articles in the tail of a
Bradford distribution, and concludes that “one may have the prejudice that items
published in the core journals of a subject are generally superior to those scattered over
journals in the tail, which is tantamount to believing that journals publishing the most
items on a topic also publish the items most worth reading, as a rule” (White, 1981, p.
50)[3]. Perhaps as responses to White’s suggestion, the proposed option is today a
standard option in the products of most online vendors.
Wallace (1987, p. 45) stated that “any search for reported uses of the results of
studies of scatter [...] will probably prove to be fruitless”, and that “there seems to be
no descriptions of the problems or benefits of actual application of such methods”. We
have found this to hold true even today. It is still not possible to locate actual reports
that describe how Bradford’s law has been applied in practical library and information
services[4]. The near absence of such reports is hard to comprehend in light of the
many suggestions for applications. It is furthermore a bit strange as G. Edward Evans
in his primer on collection management writes that:
Special libraries and information officers make good use of data generated by bibliometric
techniques in selecting and maintaining collections of the most needed serials. Bradford’s
law, Lotka’s law, Zipf’s law, and citation analysis have contributed to the effective operation
of special libraries (Evans, 2000, p. 104 [emphasis added]).
Unfortunately, Evans (2000) does not provide any details or references on this[5].
3. The concept of “subject”
Bradford’s law is explicitly about the scattering of documents on specific subjects. The
meaning of the term ”subject” (and related terms such as aboutness, topicality, and
theme) as applied in subject indexing, classification and knowledge organization, has
been investigated in LIS for about a hundred years. Among the important
contributions are Cutter (1904), Wilson (1968), Hutchins (1975, 1977, 1978), Maron
Practical
potentials of
Bradford’s law
361
(1977), Miksa (1983), Soergel (1985), and Hjørland (1992, 2001). Since Bradford
published his works, there has also been an impressive literature about Bradford’s law.
However, with few exceptions (Hjørland and Nicolaisen, 2005a; Hood and Wilson,
2001), nobody has tried to outline the consequences of different conceptions of
“subject” for Bradford’s law. The two lines of research have never really met[6]. This
lack of interest is quite peculiar as Bradford’s law precisely is about subject scattering.
However, until recently, nobody has considered what difference different conceptions
of “subject” makes for the outcome of Bradford analyses. Earlier analysts appear to
hold what Hjørland (1992, p. 173) terms a “naı
¨
ve conception of subject”, a conception
related to the philosophical position of naı
¨
ve realism. They seem to share the idea that
it is rather obvious what subjects are. Consequently, they seem to think, there is no
need to dwell much on that concept. Yet, going over the bulk of papers on Bradford’s
law and related statistical patterns, it is possible to distinguish at least three different
ways in which the concept of subject has been operationalized.
3.1 Bradford’s conception of “subject”
Let us first examine how Bradford himself perceived of the concept. Unfortunately,
Bradford never discussed the meaning of ”subject” explicitly. Consequently, we must
infer his meaning of the concept indirectly by considering how he used the word.
Bradford (1948, p. 110; 1953, p. 148) wrote under the heading “The scattering of articles
on a given subject”:
It is, therefore, necessary to examine the extent to which articles on a given subject actually
occur in periodicals devoted to quite other subjects: as, for instance, a paper on the
mechanism of the heart, contributed to the Proceedings of Physical Society, or one on genetics,
occurring in an agricultural magazine. Investigation shows that this distribution follows a
certain law, which can be deduced both theoretically from the principle of the unity of science
and practically from examination of the references.
According to this principle every scientific subject is related, more or less remotely, to
every other scientific subject.
It follows that from time to time, a periodical devoted to a special subject may contain an
article of interest from the point of view of another subject. In other words, the articles of
interest to a specialist must occur not only in the periodicals specializing on his subject, but
also, from time to time, in other periodicals, which grow in number as the relation of their
fields to that of his subject lessens and the number of articles on his subject in each periodical
diminishes.
Bradford’s empirical distribution was based on the sources indexed in four years of the
current bibliography Applied Geophysics and two and a-half years of the current
bibliography Lubrication, both prepared by the Science Library in London, of which
Bradford was the keeper. There is no discussion, however, of how papers were
assigned subject descriptors (e.g. classification codes) and how this assignment may
have influenced the actual distributions. Indirectly however, we may get a little insight
of his thinking about this issue and its consequences for his law. Bradford realized the
needs for deep indexing addressed towards specific subject areas. His library, however,
could not provide bibliographies with sufficient coverage of the relevant documents.
Because the sources were too scattered, no special library could cover all the needed
documents, and no compilatory team of a realistic size could manage to scan all the
needed sources. Instead Bradford (1953) suggested a two-step procedure. All journals
JDOC
63,3
362
and other information sources should be indexed by source, not by subject. That is: one
team should make a crude indexing of journals one by one. Then other teams of
information specialists could make specialized indexing to special purposes[7]. “On the
average, a general abstract requires two classification numbers to specify the main
subjects of the paper [...]. A special abstract, which included every substance
mentioned and every piece of apparatus described, might well need as many as twelve
classification numbers” (Bradford, 1953, p. 145).
Bradford imagined that 12 classification numbers per document would hardly be
worthwhile, as many of these numbers might never be consulted. This is in contrast to
present day information retrieval in which every word in documents may be used as
subject access point (in full text retrieval), where all references may be used as access
points (in citation indexes), and where many kinds of subject access points and
retrieval techniques may be applied. Bradford’s view reveals a mix of theoretical
considerations and practical constraints that probably are typical and harmful in the
development of general knowledge in information science.
Concerning indexing and the concept of “subject” the quotation given reveals
something about how Bradford looked at things. A comprehensive indexing should list
“every substance mentioned and every piece of apparatus described”. This is a kind of
thinking related to a listing of all “substantive” words. When he suggests that these
words should be indexed with the UDC classification, it is not the words, but the
concepts (words including synonyms and excluding homonyms) that are indexed. This
may therefore be interpreted as indexing of concepts rather than by subjects proper.
A number of subsequent studies have chosen to operationalize the concept of
subject in a similar way. A fairly recent example concerns literature on semiconductor
research (Ming-Yueh et al., 2000). Using the search command “semiconductor?. de.”, the
three authors did a Bradford analysis of twenty years of semiconductor literature in the
INSPEC database. They argued that by searching the controlled vocabulary they
would be able to control for synonyms, nearly synonyms, homographs and related
terms. This, they expected, would result in the retrieval of “most of the papers on
semiconductors” (Ming-Yueh et al., 2000, p. 492).
3.2 Title words as attributes of subjects
The view that there exists a correspondence between, for example, the title of a
document and its actual subject is not uncommon. This operationalization of “subject”
was partly employed by von Ungern-Sternberg (2000) in her Bradford analysis of
literature on the subject “information seeking”. She searched for relevant literature in
five different databases using the Dialog information system. The exact search
formulation was:
S Information (w)retrieval/de,ti or information (w)seeking/de,ti or Information
(w)behavior?/de,ti or information(w)need?/de,ti.
von Ungern-Sternberg (2000, p. 167) explains that “[t]he search was restricted to terms
in the descriptor field or in the title to ensure that the topic of the article was
information seeking”. However, as pointed out by Hjørland and Kyllesbech Nielsen
(2001), in some domains titles often express more general claims than are covered by
the documents. This presumably results in low search precision. The recall of searches
on title words has also been reported to be low. Hodges (1983) tested the effectiveness
Practical
potentials of
Bradford’s law
363
of title keywords in retrieval and found that less than 50 percent of the relevant
documents were retrieved by words in titles. The results of a study conducted by Peritz
(1984) reveals another problem. Peritz (1984) examined the frequency of
non-informative titles in LIS journals and sociology journals. She found that
non-informative titles totalled 21 percent in LIS and 15 percent in sociology. Her study
showed, moreover, that in both fields the non-informative titles were concentrated in a
few journals. Thus, a Bradford analysis of a bibliography that has been assembled by
searching journal article titles will probably be biased in favour of journals that
publish articles with informative titles.
3.3 Citations as subject markers
A third way to operationalize the concept of subject is by means of citation analysis.
Faced with the question “what files of scientific periodicals are needed in a college
library successfully to prepare the student for advanced work [in the field of
chemistry], taking into consideration also those materials necessary for the stimulation
and intellectual development of the faculty”, Gross and Gross “decided to seek an
arbitrary standard of some kind by which to measure the desirability of purchasing a
particular journal” (Gross and Gross, 1927, p. 386). In discussing potential procedures,
they rejected the method of just sitting down and compiling a list of those journals,
which were considered indispensable, because, as they wrote, “such a procedure might
prove eminently successful in certain cases, but it seems reasonably certain that often
the result would be seasoned too much by the needs, likes and dislikes of the compiler”
(Gross and Gross, 1927, p. 386). Instead, they decided to tabulate the references in a
single volume of The Journal of the American Chemical Society and subsequently to
identify the candidate journals by their citation frequencies, i.e. by the number of times
they were cited in the sample. According to Gross and Gross this tabulation could be
considered statistically and employed to predict the future needs of scientific
periodicals.
The Gross’s faith in the existence of a free and independent standard that could be
utilized for the prediction of future needs and the selection of library materials
resembles an empiricist faith in the inductive method. Pure observation of references
and their patterns were thought to make a method suitable for the selection of
excellence. A method that would lead inevitably to the identification of the paramount
of scholarly contributions: “Consideration of the method of investigation here
employed will show that we are concerned not merely with the quantity of work
published [...], but that in reality we are concerned only with the good work, the work
which have survived and which has proved of value to the investigators who followed”
(Gross, 1927, p. 641). A number of subsequent studies and Bradford analyses have
operationalized the concept of subject in similar ways (e.g., Coile, 1952 [subject:
Electrical Engineering]; Cave, 1963 [Tropical Agriculture]; Meadows, 1967
[Astronomy]; Donohue, 1972 [Information Science]; Yeon-Kyoung, 1994
[Classification Systems]).
The efficiency of this approach is determined by how well citing documents
identifies and cites relevant sources in their reference lists. As pointed out by Hjørland
and Kyllesbech Nielsen (2001), the method presupposes that scientific literature is
neither unrelated to other research and literature nor simply redundant. It also
presupposes that authors do not cite on purely formal or presentational grounds. But
JDOC
63,3
364
most importantly, it presupposes that authors are not biased when selecting their
sources but give even consideration to papers that argue both for and against their own
view. This last assumption contradicts both qualitative and quantitative research on
the citing behavior of scientists. Delamont (1989) conducted a qualitative study of
citation patterns between schools of researchers studying social mobility in Britain. In
her study she was able to demonstrate systematic neglect by each school of the work of
the others. Her article is loaded with examples showing the failure of leading scholars
to address the work of others. She concludes, in fact, that this is the most striking
feature on the literature on social mobility. Inspired by Hjørland (2002), Nicolaisen
(2004) conducted a structural bibliometric analysis of citations between 16 leading
psychological journals representing four different research traditions. The results of
his empirical study indicate that authors from different research traditions tend to use
and cite sources that share their own basic views while ignoring contradicting
perspectives. As observed by Delamont (1989, p. 335), “[t]he most devastating way of
demonstrating that another scholar is not part of the in-crowd is to leave them out of
debate all together – to render them invisible”. And as noted by Meadows (1974, p. 45),
“[i]f incorrect results stand in the way of the further development of a subject, or if they
contradict work in which someone else has a vested interest, then it may become
necessary to launch a frontal attack [...]. Otherwise, it generally takes less time and
energy to bypass erroneous material, and simply allow it to fade into obscurity”. Thus,
disagreement has probably very little influence on the distribution of references and
citations. As noted by Cole and Cole (1974, p. 33), “[p]apers which are trivial and receive
critical citations will not accumulate large numbers of citations”. This is, of course, a
generalization. Garfield (1989, p. 10), however, claims it is one he has found to apply in
most cases. Yet there are important exceptions. The famous article on cold fusion from
1989, for instance. Meadows (1998) found that this article had been cited several
hundred times, mostly by researchers who disputed its results. Reiterating the
essentials of his 1974 argument, Meadows (1998, p. 90) concluded that “[s]uch citation
depends on the importance of the topic: questionable articles dealing with less
important topics are likely to be ignored rather than cited”.
4. Neutrality
We have previously identified three kinds of scattering[8] (Hjørland and Nicolaisen,
2005a, p. 103):
(1) Lexical scattering. The scattering of words in texts and in collections of texts.
(2) Semantic scattering. The scattering of concepts in texts and in collections of
texts.
(3) Subject scattering. The scattering of items useful to a given task or problem.
These three kinds of scattering are not completely independent of each other. Previous
research has shown that there are some internal relations between them – they do
overlap to some extent depending on the domain of study. The overlap between
semantic scattering and subject scattering (operationalized by citation relations) has,
however, been reported to be quite low. Pao (1984) was among the first to suggest that
descriptor and citation retrieval are likely to produce quite different search results. She
examined the results of 10 searches (on a medical subject) using both descriptors and
citations for retrieving relevant documents from a proprietary database. The result of
Practical
potentials of
Bradford’s law
365
her study demonstrated a “surprisingly low overlap” (Pao, 1984, p. 136). In a later
study, Pao and Fu (1985) found that 12 percent of a document set on sulfur dioxide air
pollution was retrievable both from MEDLINE and by citation searches. McCain (1989)
reported a comparable low overlap (10 percent) in her study of descriptor retrieval
versus citation retrieval in the field of biomedicine. Green (2000) compared citation
searching with the use of standard bibliographic tools in the humanities and found an
even lower overlap. Only 5 percent of the retrieved items were found by both search
tactics. The overlap between lexical scattering and semantic scattering is probably
somewhat higher. Miller (1971) compared the retrieval efficiency of MEDLARS titles
and index terms and found that title words retrieved close to 40 percent of the relevant
documents retrieved by index terms, and that index terms retrieved close to 55 percent
of the relevant documents retrieved by title words. Olive et al. (1973) did a similar study
of retrieval effectiveness in Nuclear Science Abstracts. They found that both index term
searching and title words searching missed many relevant documents retrieved by the
other. Index terms retrieved 58 percent of the relevant documents retrieved by title
words. The converse relationship was only 51 percent.
What difference do different conceptions of “subject” have for the outcome of
Bradford analyses? It seems reasonable to assume that depending on how the concept
of subject is being operationalized (e.g., descriptors, title words, references), different
results will emerge. Yet, it need not be so. Relevant documents retrieved by any search
procedure (e.g., by descriptors, title words or references) could be distributed equally
among the same set of journals. If this is the case, different conceptions of “subject”
should have little or no influence on the results of Bradford analyses. A Bradford
analysis conducted using title words for searching relevant documents would thus
identify the same core journals (journals in the first Bradford zone) as other analyses
based on descriptor searches or citation searches (etc.). In order to test our assumption
that the outcome of any Bradford analysis will be heavily affected by the way the
subject is operationalized, we conducted three Bradford analyses of two different
subjects:
(1) Anorexia nervosa; and
(2) Virology.
The methods employed for the three Bradford analyses and the results of the same are
described below.
4.1 Anorexia nervosa
Anorexia nervosa is an eating disorder characterized by voluntary starvation. It is a
complex disease that probably involves both psychological and physiological
components. However, the causes of anorexia nervosa are not entirely agreed upon. We
decided to take a closer look at the psychological side of the disease and chose the
database PsycInfo and the journal International Journal of Eating Disorders as data
collections for three Bradford analyses:
(1) Articles published during the years 2000-2004 and retrieved in PsycInfo by a
search for “anorexia” in the title and/or abstract fields.
(2) Articles published in the same period and retrieved in PsycInfo by a search for
“anorexia nervosa” in the descriptor field.
JDOC
63,3
366
(3) Items cited in articles published in International Journal of Eating Disorders
during the years 2000-2004. The search was limited to citing articles with the
term “anorexia” in the title and/or abstract.
Table I lists the journals in the first Bradford zone. The table reveals that the three
different operationalizations of “anorexia nervosa” identify different collections of
journals in the first zone. As seen in Table II, the total overlap is just 5.9 percent.
4.3 Virology
Virology is the study of viruses and their properties. We chose the database BIOSIS
Previews and the journal Journal of Virology as data collections for three Bradford
analyses:
(1) Articles published during the years 2000-2004 and retrieved in BIOSIS Previews
by a search for “virology” in the title and/or abstract fields.
(2) Articles published in the same period and retrieved in BIOSIS Previews by a
search for “virology” in the descriptor field.
(3) Items cited in articles published in Journal of Virology during the years
2000-2004.
Table III lists the journals in the first Bradford zone. The table reveals that the three
different conceptions of “virology” identify different collections of journals in the first
zone. As seen in Table IV, the total overlap is just 17.6 percent.
These results clearly show that Bradford analyses are heavily affected by the way
the subject is operationalized. A Bradford analysis conducted using title and abstract
words for searching relevant documents does not identify the same core journals
(journals in the first Bradford zone) as other analyses based on descriptor searches or
citation searches. The three different operationalizations lead, in fact, to the
identification of very different collections of core-journals.
5. Objectivity
Andersen (2000) presents the results of a survey of 788 Danish researchers from the
social sciences. The researchers had been asked to assess the most influential
researchers and journals of their fields. The results display a pluralistic picture and
only a moderate degree of consensus among the researchers. Except for one
exceptional case (legal science), the percentage of researchers mentioning the same
journal as one of the three most important ones was found to be 25 percent or less.
Andersen (2000, p. 689) thus concludes that “although a set of common journals can
function as a medium of communication, thus making visible the reputation of research
results and researchers, this function is weak, at least in the Danish social sciences”.
Yet, the same pluralistic picture and moderate degree of consensus may possibly
characterize much larger parts of the social sciences. The cognitive school has
dominated American psychology (and thus also international psychology) since the
early 1970s. Before that time behaviorism was the dominant school in psychology
(Robins et al., 1999). However, historians of psychology have found that cognitive
approaches may be tracked long back in the history of psychology. Greenwood (1991),
for instance, identified sources of cognitive psychology in the hey-days of behaviorism,
often in more remote psychological journals. It is reasonable to expect, that if
Practical
potentials of
Bradford’s law
367
Title and abstract Descriptor References
American Journal of Psychiatry Eating and Weight Disorders Acta Psychiatrica Scandinavica
Eating and Weight Disorders Eating Behaviors American Journal of Clinical Nutrition
Eating Behaviors European Eating Disorders Review American Journal of Psychiatry
Eating Disorders: The Journal of Treatment International Journal of Eating Disorders Archives of General Psychiatry
European Eating Disorders Review International Journal of Law and Psychiatry Biological Psychiatry
International Journal of Eating Disorders Molecular Psychiatry British Journal of Psychiatry
International Journal of Law and Psychiatry Psychiatria Polska British Medical Journal
Molecular Psychiatry Clinical Psychology Review
Physiology & Behavior Comprehensive Psychiatry
Psychiatria Polska European Eating Disorders Review
Seishin Igaku (Clinical Psychiatry) International Journal of Eating Disorders
JAMA – Journal of The American Medical
Association
Journal of Abnormal Psychology
Journal of Clinical Endocrinology and Metabolism
Journal of Clinical Psychiatry
Journal of Consulting and Clinical Psychology
Journal of Nervous and Mental Disease
Journal of Psychiatric Research
Journal of Psychosomatic Research
Journal of The American Academy of Child and
Adolescent Psychiatry
Lancet
New England Journal of Medicine
Psychiatric Clinics of North America
Psychiatry Research – Neuroimaging
Psychological Medicine
Psychosomatic Medicine
Table I.
Journals in the first
Bradford zone according
to three
operationalizations of the
subject (anorexia
nervosa)
JDOC
63,3
368
Bradford’s law had been applied to select psychology journals to libraries around 1930,
then the result would have been that journals with an exclusive behavioral orientation
would have been too strongly represented and that journals more open towards, for
example, the cognitive approach, would have been too weakly represented.
Consequently, there is no reason to believe that the assumptions about universalism
and objectivity underlying the application of Bradford’s law can be combined with the
demands on pluralism, which may be expected from libraries and information systems.
In other words: The application of Bradford’s law to the selection of information
sources is probably not an objective method, but may possibly turn out to function
discriminatorily against minority views[9]. To test whether this is the case, we
searched Social SciSearch and identified four collections of data. The first collection
consisted of 536 articles from 2004 that cited at least one of four leading
psychoanalytical journals[10]. The second collection consisted of 2,784 articles from
2004 that cited at least one of four leading behaviorist journals[11]. The third collection
%
Title and abstract/descriptor 63.6
Title and abstract/references 8.8
Descriptor/references 6.5
Title and abstract/descriptor/references 5.9
Table II.
Overlaps (anorexia
nervosa)
Title and abstract Descriptor References
Journal of General Virology Archives of Virology Cell
Journal of Medical Virology Journal of General Virology Embo Journal
Journal of Virological Methods Journal of Medical Virology Journal of Biological Chemistry
Journal of Virology Journal of Virology Journal of Experimental Medicine
Virus Research Virology Journal of General Virology
Virology Journal of Immunology
Journal of Virology
Molecular and Cellular Biology
Nature
Nucleic Acids Research
Proceedings of The National Academy
of Sciences of The United States of
America
Science
Virology
Table III.
Journals in the first
Bradford zone according
to three
operationalizations of the
subject (virology)
%
Title and abstract/descriptor 57.1
Title and abstract/references 18.8
Descriptor/references 21.0
Title and abstract/descriptor/references 17.6
Table IV.
Overlaps (virology)
Practical
potentials of
Bradford’s law
369
consisted of 3.868 articles from 2004 that cited at least one of four leading cognitive
journals[12]. The fourth collection consisted of 6,987 articles from 2004 that cited at
least one of these 12 journals. We did Bradford analyses on these four collections and
thus identified four collections of core journals (journals in the first Bradford zone) (see
Table V).
The results of the four Bradford analyses reveal that Bradford’s law does function
discriminatorily against minority views. The Bradford analysis of all 12 leading
journals identifies ALL of the cognitive core-journals, but only 45 percent of the
behavioral core-journals and just 50 percent of the psychoanalytic. One could ask[13]
whether this is really “discrimination”. Do not the results simply reflect “the
contemporary view”? This is, for instance, close to how Cole and Cole (1973, p. 24) used
to argue for their utilization of citation analysis:
Current citations are not a measure of the absolute quality of work, they are an adequate
measure of the quality of work socially defined.
Yet, we should not forget that our choices always have consequences. Thus, if one
chooses to rely solely on the result of a Bradford analysis and thus to go exclusively
with the majority view when selecting journals for a library collection, it is definitely
an act of discrimination against the overlooked and disregarded minority views.
6. Conclusion
In this paper we have put forward serious arguments against the received view on
Bradford’s law. The fact that it is difficult to find any examples of its actual use in
practice may be an indication that such problems have intuitively been foreseen.
What is a subject for one person need not be the same subject for another. The best
way to generalize views about subjects is probably to consider different theoretical
views or epistemologies regarding that subject. A pure mechanical view of selection
must consequently be replaced by a reflective view in which the selector must justify
the selection. Hjørland (1992) proposed that “subjects” are the informative or
epistemological potentials of documents. The purpose of subject analysis and indexing
is to make some documents retrievable in relation to some purposes. A subject analysis
is always preliminary because you cannot in advance predict what kinds of use will be
made of a document. The whole issue is theory-dependent. For example, nobody could
predict that the subject “energy resource” was adequate to uranium before
radioactivity was discovered. In the same way will the potential usefulness of
documents be dependent on theoretical developments in the user community. The
implication is that dynamic kinds of subject assignments[14] have better possibilities
to adapt to changing needs by the users compared to fixed indexing systems like
descriptors and library classifications.
The implication for Bradford analyses is that subject analysis is very relative and
depending on subject knowledge. No kind of mechanical operationalizations of subject
analysis (say: by title, descriptor, or citations) can be said to be the best one in all cases.
What is the best way to operationalize subject analysis is simply depending on the
specific query as well as the standard of the available subject access points.
This insight introduces a major subjective element in Bradford analyses. The
claimed objectivity of the method is thus based on a subjective factor. Because the
JDOC
63,3
370
Psychoanalytic core-journals Behavioral core-journals Cognitive core-journals Psychological core-journals
International Journal of
Psychoanalysis
Addictive Behaviors Applied Cognitive Psychology Acta Psychologica
Journal of The American
Psychoanalytic Association
Behavior Analyst Behavior Research Methods
Instruments & Computers
Applied Cognitive Psychology
Psyche-Zeitschrift fu
¨
r Psychoanalyse
und Ihre Anwendungen
Behavior Modification
Brain and Cognition
Behavior Research Methods
Instruments & Computers
Psychoanalytic Inquiry
Behavior Therapy
Brain and Language Behavior Therapy
Psychoanalytic Psychology
Behavioral Interventions Cognition Behaviour Research and Therapy
Psychoanalytic Quarterly
Behaviour Research and Therapy European Journal of Cognitive
Psychology
Behavioural Processes
Behavioural and Cognitive
Psychotherapy Journal of Cognitive Neuroscience
Brain and Cognition
Behavioural Processes
Journal of Experimental
Psychology-Human Perception and
performance
Brain and Language
Cognition & Emotion
Journal of Experimental Psychology
–Learning, Memory and Cognition
Child Development
Cognitive and Behavioral Practice
Journal of Memory and Language
Cognition
Cognitive Therapy and Research
Memory
Cognition & Emotion
Depression and Anxiety
Memory & Cognition
Cognitive and Behavioral Practice
International Journal of Eating
Disorders
Neuropsychologia
Cognitive Neuropsychology
Journal of Abnormal Psychology
Neuropsychology
Cognitive Science
Journal of Affective Disorders
Perception
Cognitive Therapy and Research
Journal of Anxiety Disorders
Perception & Psychophysics
Developmental Psychology
Journal of Applied Behavior Analysis
Perceptual and Motor Skills
Developmental Science
Journal of Behavior Therapy and
Experimental Psychiatry
Psychological Science
European Journal of Cognitive
Psychology
Journal of Child Psychology and
Psychiatry
Psychology and Aging
International Journal of
Psychoanalysis
Journal of Clinical Child and
Adolescent Psychology Psychonomic Bulletin & Review
Journal of Abnormal Psychology
Journal of Clinical Psychiatry Quarterly Journal of Experimental
Psychology Section A –Human
Experimental Psychology
Journal of Anxiety Disorders
Journal of Clinical Psychology
Visual Cognition
Journal of Applied Behavior Analysis
Journal of Consulting and Clinical
Psychology
Journal of Child Psychology and
Psychiatry
Journal of Nervous and Mental Disease
Journal of Clinical Psychology
Journal of Positive Behavior
Interventions
Journal of Cognitive Neuroscience
Journal of Psychopathology and
Behavioral Assessment
Journal of Consulting and Clinical
Psychology
Journal of Psychosomatic Research
Journal of Experimental Child
Psychology
(continued)
Table V.
Journals in the first
Bradford zone according
to three
operationalizations of the
subject (Psychology)
Practical
potentials of
Bradford’s law
371
Psychoanalytic core-journals Behavioral core-journals Cognitive core-journals Psychological core-journals
Journal of The American Academy of
Child and Adolescent Psychiatry
Journal of Experimental Psychology –
Human Perception and Performance
Journal of the Experimental Analysis
of Behavior
Journal of Experimental Psychology –
Learning Memory and Cognition
Journal of Traumatic Stress Journal of Memory and Language
Learning & Behavior Journal of Personality and Social
PsychologyMemory
Journal of The American
Psychoanalytic Association
Pain
Journal of the Experimental Analysis
of Behavior
Personality and Individual Differences
Memory
Psych
Memory & Cognition
Psychiatry Research
Neuroimage
Psychological Record
Neuropsychologia
Psychological Reports
Neuropsychology
Research in Developmental Disabilities
Perception
Perception & Psychophysics
Perceptual and Motor Skills
Personality and Individual Differences
Psychoanalytic Psychology
Psychological Record
Psychological Reports
Psychological Science
Psychology and Aging
Psychonomic Bulletin & Review
Quarterly Journal of Experimental
Psychology Section A – Human
Experimental Psychology
Visual Cognition
Table V.
JDOC
63,3
372
discussion of this subjective factor has never been raised in relation to Bradford
analyses, their claimed neutrality and objectivity are based on a myth.
One common practice in research libraries is to have subject specialists selecting the
journals. This praxis represents a view that is based on assumptions that differ from
the use of Bradford analyses. Still, one may ask how research in LIS could contribute to
make the selection better motivated? Our answer is that investigations such as the one
by Andersen (2000) that study researchers’ attitudes toward journals and compare
patterns across disciplines and across national borders and compare such opinions
with the rankings from citation databases provide the most valuable information for
the selection of journals which can be obtained from research.
Notes
1. Bradford believed n to be constant in the different zones (n
1
¼ n
2
¼ n); Results reported by
Rao (1998) indicates, however, that Bradford’s assumption was wrong: Bradford multipliers
vary from zone to zone.
2. The received view on Bradford’s law is the view put forward by the majority of textbooks
(see, e.g. Evans, 2000; Nisonger, 1998).
3. This prejudice is probably quite common. Sandstrom (2001, p. 584), for instance, argues:
Knowing how the core is constructed and integrated with other research concerns makes it
easier for scholars to track down necessary information.
4. One exception is the ISI databases. The ISI journal selection process is claimed to be based in
part on Bradford analyses. Yet, the relative contribution of this method is not reported
(http://scientific.thomson.com/knowtrend/essays/selectionofmaterial/journalselection/).
5. One of the referees argued that the reason why it is hard to find detailed applications of
Bradford’s law in practice is because it has long been recognized as great fun in theory, but
not to be taken too seriously by practitioners. This, however, is not the impression one
receives from reading the literature on Bradford’s law. Besides the already mentioned
sources, one needs only consult last years’ volume of Scientometrics to see that the
contemporary view on Bradford’s law is actually the received view. The article by Kumar
Patra and Chand (2005, p. 590) argues, for instance, that [b]ecause of increasing cost of
information resources in today’s library and information centers, a typical Bradford analysis
can suggest the journals to be procured in a library collection. In fact, the authors claim (p.
583) to have used Bradford’s law to identify the core journals of Indian Biotechnology. Thus,
we find it hard to agree with the anonymous referee about Bradford’s law having long been
recognized as being of limited applicability in practice. The same referee argued that
Bradford’s law was devised in the context of science, and is most useful in that context. The
main reason is that a given area of science nearly always has a theoretical framework
accepted by the great majority of its practitioners. The same is not generally true either of
the social sciences or of the humanities. However, Bradford’s law has been examined in both
the social sciences and the humanities (e.g., Donohue, 1972; Heine, 1998; von
Ungern-Sternberg, 2000; Yeon-Kyoung, 1994). Bradford’s law in relation to questions
concerning theoretical frameworks accepted by the majority has, as far as we know, never
been examined or discussed. The view that science is governed by degrees of theoretical
consensus is, by the way, a post-positivist view raised – in different ways – by, for instance,
Thomas Kuhn and Karl Popper. Such epistemological discussions are in our view very
important but have unfortunately been rather neglected in LIS. That is actually partly why
we launched The Epistemological Lifeboat (Hjørland and Nicolaisen, 2005b).
6. Umsta
¨
tter (2005) has claimed in a German e-only journal published at the Department of
Library and Information Science/Humboldt-University Berlin by a student editorial team
Practical
potentials of
Bradford’s law
373
that we (Hjørland and Nicolaisen, 2005a) are wrong: “Am Schluss sei noch darauf
hingewiesen, dass die Aussage” nobody have thus far tried to outline the consequences of
different conceptions of “subjects” for Bradford’s law “die wichtige Diskussion vieler
Rechercheure u
¨
ber Relevanz, Precision, Recall, Noise und Pertinenz vo
¨
llig vernachla
¨
ssigt”.
However, Umsta
¨
tter failed to document his claim. He failed to produce proper evidence that
these concepts have been regarded in relation to consequences for Bradford’s law. We
seriously doubt the existence of such documentation.
7. In our opinion Bradford’s suggestion is unrealistic and based on a too simplified view of
subject matter. It is not possible to make screening procedures of this kind. It is impossible to
screen documents in relation to user needs, and it is not possible to screen user requirements
in relation to subject classes or descriptors. Based on the second author’s experience as a
practicing academic librarian, our view is that such screenings are most often harmful (see
Hjørland, 1997, p. 155 n4). However, we know of no investigations that have made serious
inquiries into this problem. The problem falls outside the scope of this article.
8. The idea about different kinds of scattering occurred when the second author served as
referee for a paper by Hood and Wilson (2001). This paper examines – in the tradition from
Bradford (1934) – the distribution of bibliographic records in online bibliographic databases
using 14 different search topics, searched in the Dialog information system. The second
author’s suggestion to the two authors was that these 14 questions represented different
kinds of questions and thus different kinds of scattering. This was subsequently
acknowledged by Hood and Wilson (2001, p. 1253-4).
9. This is especially interesting in the case reported because this view later developed to a
majority view. If journals open to the cognitive view had been deselected due to the
application of Bradford’s law, this could perhaps have made it more difficult for this view to
develop.
10. International Journal of Psychoanalysis, Psychoanalytic Quarterly, Journal of the American
Psychoanalytic Association and Contemporary Psychoanalysis was identified as leading
psychoanalytic journals by Robins et al. (1999, p. 121) in their empirical analysis of trends in
psychology.
11. Journal of the Experimental Analysis of Behavior, Behavior Research and Therapy, Journal of
Applied Behavior Analysis and Behaviour Therapy was identified as leading behavioral
journals by Robins et al. (1999, p. 121) in their empirical analysis of trends in psychology.
12. Cognitive Psychology, Cognition, Memory and Cognition and Journal of Experimental
Psychology: Learning, Memory, and Cognition was identified as leading cognitive journals
by Robins et al. (1999, p. 121) in their empirical analysis of trends in psychology.
13. As one of the anonymous referees in fact did.
14. Dynamic assignments of subject indicators are possible using bibliometric methods like, for
example, co-citation analysis. Such methods have other drawbacks not to be discussed here.
The point at this place is that kinds of dynamic subject assignments – other things being
equal – are to be preferred compared to static kinds of subject assignments.
References
Andersen, H. (2000), “Influence and reputation in the social sciences – how much do researchers
agree?”, Journal of Documentation, Vol. 56 No. 6, pp. 674-92.
Bradford, S.C. (1934), “Sources of information on specific subjects”, Engineering, Vol. 26, pp. 85-6.
Bradford, S.C. (1948), Documentation, Crosby Lockwood, London.
Bradford, S.C. (1953), Documentation, 2nd ed., Crosby Lockwood, London.
Brookes, B.C. (1969), “Bradford’s law and the bibliography of science”, Nature, Vol. 224, pp. 953-6.
JDOC
63,3
374
Cave, R. (1963), “Tropical agriculture literature citations”, Quarterly Bulletin of the International
Association of Agricultural Librarians and Documentalists, Vol. 8 No. 3, pp. 163-9.
Coile, R.C. (1952), “Periodical literature for electrical engineers”, Journal of Documentation, Vol. 8
No. 4, pp. 209-26.
Cole, J.R. and Cole, S. (1973), Social Stratification in Science, University of Chicago Press, Chicago,
IL.
Cole, J.R. and Cole, S. (1974), “Citation analysis”, Science, Vol. 183, pp. 28-33.
Cutter, C.A. (1904), Rules for a Dictionary Catalog, Government Printing Office, Washington, DC.
Delamont, S. (1989), “Citation and social mobility research: self defeating behaviour?”,
Sociological Review, Vol. 37, pp. 332-7.
Diodato, V. (1994), Dictionary of Bibliometrics, Haworth Press, Binghamton, NY.
Donohue, J.C. (1972), “A bibliometric analysis of certain information science literature”, Journal of
the American Society for Information Science, Vol. 23 No. 5, pp. 313-17.
Evans, G.E. (2000), Developing Library and Information Center Collections, 4th ed., Libraries
Unlimited, Englewood, CO.
Garfield, E. (1989), “Of hot papers and critical acclaim”, The Scientist, Vol. 3 No. 4, p. 10.
Green, R. (2000), “Locating sources in humanities scholarship: the efficacy of following
bibliographic references”, Library Quarterly, Vol. 70 No. 2, pp. 201-29.
Greenwood, J.K. (1991), Relations and Representations: An Introduction to the Philosophy of
Social Psychological Science, Chapman & Hall, New York, NY.
Gross, P.L.K. (1927), “Fundamental science and war”, Science, Vol. 66, pp. 640-5.
Gross, P.L.K. and Gross, E.M. (1927), “College libraries and chemical education”, Science, Vol. 66,
pp. 385-9.
Heine, M.H. (1998), “Bradford ranking conventions and their application to a growing literature”,
Journal of Documentation, Vol. 54 No. 3, pp. 303-31.
Hjørland, B. (1992), “The concept of subject in information science”, Journal of Documentation,
Vol. 48 No. 2, pp. 172-200.
Hjørland, B. (1997), Information Seeking and Subject Representation. An Activity-theoretical
approach to Information Science, Greenwood Press, Westport, CT and London.
Hjørland, B. (2001), “Towards a theory of aboutness, subject, topicality, theme, domain, field,
content... and relevance”, Journal of the American Society for Information Science and
Technology, Vol. 52 No. 9, pp. 774-8.
Hjørland, B. (2002), “Epistemology and the socio-cognitive perspective in information science”,
Journal of the American Society of Information Science and Technology, Vol. 53 No. 4,
pp. 257-70.
Hjørland, B. and Kyllesbech Nielsen, L. (2001), “Subject access points in electronic retrieval”,
Annual Review of Information Science and Technology, Vol. 35, pp. 249-98.
Hjørland, B. and Nicolaisen, J. (2005a), “Bradford’s law of scattering: ambiguities in the concept
of subject”, Proceedings of the 5th International Conference on Conceptions of Library and
Information Sciences, pp. 96-106.
Hjørland, B. and Nicolaisen, J. (Eds) (2005b), The Epistemological Lifeboat, available at: www.db.
dk/jni/lifeboat/home.htm.
Hodges, P.R. (1983), “Keyword in title indexes”, Special Libraries, Vol. 74 No. 1, pp. 56-60.
Practical
potentials of
Bradford’s law
375
Hood, W. and Wilson, C.S. (2001), “The scatter of documents over databases in different subject
domains: how many databases are needed?”, Journal of the American Society for
Information Science and Technology, Vol. 52 No. 14, pp. 1242-54.
Hutchins, W.J. (1975), Languages of Indexing and Classification: A Linguistic Study of Structures
and Functions, Peter Peregrinus, London.
Hutchins, W.J. (1977), “On the problem of aboutness in document analysis”, Journal of
Informatics, Vol. 1, pp. 17-35.
Hutchins, W.J. (1978), “The concept of aboutness in subject indexing”, Aslib Proceedings, Vol. 30,
pp. 172-81.
Kumar Patra, S. and Chand, P. (2005), “Biotechnology research profile of India”, Scientometrics,
Vol. 63 No. 3, pp. 583-97.
McCain, K.W. (1989), “Descriptor and citation retrieval in the Medical Behavioral Sciences
literature: retrieval overlaps and novelty distribution”, Journal of the American Society for
Information Science, Vol. 40 No. 2, pp. 110-14.
Maron, M.E. (1977), “On indexing, retrieval and the meaning of about”, Journal of the American
Society for Information Science, Vol. 28, pp. 38-43.
Meadows, A.J. (1967), “The citation characteristics of astronomical research literature”, Journal of
Documentation, Vol. 23, pp. 28-33.
Meadows, A.J. (1974), Communication in Science, Butterworth, London.
Meadows, A.J. (1998), Communicating Research, Academic Press, San Diego, CA.
Miksa, F. (1983), The Subject in the Dictionary Catalog from Cutter to the Present, American
Library Association, Chicago, IL.
Miller, W.L. (1971), “The efficiency of MEDLARS titles for retrieval”, Journal of the American
Society for Information Science, Vol. 22, pp. 318-21.
Ming-Yueh, T., Shiow-Jen, J. and Sheau-Shin, M. (2000), “A bibliometric study of semiconductor
literature, 1978-1997”, Scientometrics, Vol. 49 No. 3, pp. 491-509.
Nicolaisen, J. (2004), “Social behavior and scientific practice – missing pieces of the citation
puzzle”, PhD thesis, Royal School of Library and Information Science, Copenhagen.
Nisonger, T.E. (1998), Management of Serials in Libraries, Libraries Unlimited, Englewood, CO.
Olive, G., Terry, J.E. and Datta, S. (1973), “Studies to compare retrieval using titles with that
using index terms: SDI from Nuclear Science Abstracts”, Journal of Documentation, Vol. 29
No. 2, pp. 169-91.
Pao, M.L. (1984), “Semantic and pragmatic retrieval”, Proceedings of the 47th ASIS Annual
Meeting, pp. 134-6.
Pao, M.L. and Fu, T.T.W. (1985), “Titles retrieved from Medline and from citation relationships”,
Proceedings of the 48th ASIS Annual Meeting, pp. 120-3.
Peritz, B.C. (1984), “On the informativeness of titles”, International Classification, Vol. 11 No. 2,
pp. 87-9.
Rao, I.K.R. (1998), “An analysis of Bradford multipliers and a model to explain law of scattering”,
Scientometrics, Vol. 41 Nos 1/2, pp. 93-100.
Robins, R.W., Gosling, S.D. and Craik, K.H. (1999), “An empirical analysis of trends in
psychology”, American Psychologist, Vol. 54 No. 2, pp. 117-28.
Sandstrom, P.E. (2001), “Scholarly communication as a socioecological system”, Scientometrics,
Vol. 51 No. 3, pp. 573-605.
JDOC
63,3
376
Soergel, D. (1985), Organizing Information: Principles of Data Base and Retrieval Systems,
Academic Press, Orlando, FL.
Umsta
¨
tter, W. (2005), “Anmerkungen zu Birger Hjørland und Jeppe Nicolaisen” (Bradford’s Law
of Scattering: Ambiguities in the Concept of Subject), Libreas. Library Ideas, Vol. 3 No. 1,
available at: www.ib.hu-berlin.de/ , libreas/libreas_neu/ausgabe3/008ums.htm; www.ib.
hu-berlin.de/ , libreas/libreas_neu/ausgabe3/pdf/008ums.pdf
von Ungern-Sternberg, S. (2000), “Bradford’s law in the context of information provision”,
Scientometrics, Vol. 49 No. 1, pp. 161-86.
Wallace, D.P. (1987), “A solution in search of a problem: bibliometric and libraries”, Library
Journal, Vol. 1, pp. 43-7.
White, H.D. (1981), “Bradfordizing search output: how it would help online users”, Online Review,
Vol. 5, pp. 47-54.
Wilson, P. (1968), Two Kinds of Power: An Essay on Bibliographical Control, University of
California Press, Berkeley, CA.
Yeon-Kyoung, C. (1994), “Core international journals of classification systems: an application of
Bradford’s law”, Knowledge Organization, Vol. 21 No. 2, pp. 75-83.
Corresponding author
Jeppe Nicolaisen can be contacted at: jni@db.dk
Practical
potentials of
Bradford’s law
377
To purchase reprints of this article please e-mail: reprints@emeraldinsight.com
Or visit our web site for further details: www.emeraldinsight.com/reprints
