Library and information science according to the citing pattern of students: a bibliometric study

Introduction

This paper uses bibliometric techniques to conceptualize library and information science according to the information use of students, instead of the information use of researchers. It does so by using an institutional repository of master theses as the main data source, an approach that is quite uncommon in bibliometrics.

There are several good reasons for taking such an alternative approach. The first one is that drawing from alternative data sources to investigate library and information science help to even out an unbalanced account of the field. Many bibliometric studies of library and information science has been performed over the years, almost without exceptions based on research papers from Web of Science (see for example Persson (1994), White and McCain (1998), Åström (2006), and Lariviére et al (2012)). The findings of these have influenced how the field is conceived in terms of major areas of specialization, influential thinkers, and so on. By relying exclusively on one kind of data source there is a risk of underestimating the impact of research not well represented in it. Moed (2005) points to the social sciences and especially the humanities (two areas often associated with library and information science) as areas that are only moderately covered by Web of Science, indicating that rankings and mappings produced from such data will miss important components. Chang and Huang (2012) describes library and information science as a multifaceted and interdisciplinary field, mixing both qualitative and quantitative methods and closely connected to general sciences, computer science, business/management, education, and sociology. Åström (2006, p. 59), using the concept of fragmented adhocracy, describes it as a ‘field with high levels of task uncertainty and low mutual dependence, where research is weakly coordinated, providing plenty of opportunities for pursuing personal research agendas’; and furthermore lacking in standardization both in standards for evaluating research and in the communication system. It is unlikely that all aspects of such a diverse and fragmented field can be captured by studying only the research papers indexed in Web of Science or Scopus. Thus, using student master theses as data source in a bibliometric analysis might illuminate components of the field that previously might have gone unnoticed.

A second reason is to shift the bibliometric perspective from investigating library and information science as a field of research to also investigate it as a field of education. Bibliometrics and scientometrics has had a long tradition of centring their analysis on the world of researchers, by drawing on data harvested from citation indexes. Often the inclusion of publications in such indexes is based on notions of impact or importance for the scientific community. Therefore, analyses based on data from such indexes can only map fields of research. As noted by Åström (2006, p. 60) the research performed in library and information science is not only directed toward other researchers, but the field is also directly impacted by professional practice; for example, ‘by contributions to the research literature, by being an important audience group and by providing research funds’. Several scholars (Åström, 2006; Booth, 2003; Schlögl and Stock, 2008) also point to the role played in producing literature used in the education of librarians or information specialists, as well as literature providing a sound theoretical underpinning for practices and services. This kind of practice or education-oriented research can only be analysed bibliometrically by using alternative data such as student master theses (or similar), and what such an analysis give is a bibliometric perspective of library and information science as a field of education. An analysis of such data would therefore uncover research and researchers important for the professional activities and education of librarians.

Numerous studies exist that, for different purposes, analyse bibliographies produced by students (Davis, 2002, 2003; Davis and Cohen, 2001; Hurst and Leonard, 2007; Jarneving, 2001; Kushkowski et al., 2003; Pilerot, 2007; Rafferty, 2013). However, only a small subset of these analyse master theses from library and information science (Iseborn and Swartz, 2004; Ohlsson, 2004; Oppenheim and Smith, 2001; Waldh, 2013), and no study has yet to been published that utilize more advanced techniques such as bibliographic coupling to the end of conceptualizing the field as a field of education. Thus, this paper fills a clear gap in the literature.

The first part of this paper constitutes an investigation into the information use of Swedish master students in library and information science. This is studied by classifying and analysing the documents referenced to in the master theses. The second part of this paper focus on exploring the different specializations of Swedish library and information science as represented by the master theses themselves. This is done by automatically grouping theses of similar focus by applying: (1) bibliographic coupling as a measure of similarity between them; (2) cluster analysis for assigning group membership; (3) classifying the clusters by combining manual inspection with the use of Term Frequency–Inverse Document Frequency of theses keywords.

Data Collection

By (A) extracting, (B) standardizing and (C) classifying reference data from 210 master theses (published in 2008 and 2009 at The Swedish School of Library and Information Science, University of Borås) harvested from the Borås Academic Digital Archive repository a citation index was constructed that would meet the criteria of an in-depth bibliometric analysis (a complete list of the theses included in the data set is given in Appendix A).

Method

Basic bibliometric indicators

Using Excel and SPSS, the core authors (i.e. the most cited authors) and the core documents (i.e. the most cited documents) was calculated along with basic bibliometric indicators such as the Price Index. According to Haustein (2012, p. 35) analysing the cited sources in such ways ‘can shed light on the degree of the diversity regarding the […] knowledge base’ of a field or journal.

The Price index, introduced by Derek de Solla Price (1970) as a way to quantify the hardness or softness of a subject area, is defined as the share of references not older than five years. It measures the currentness of the literature used by the students in the master theses.

Bibliographic coupling

Bibliographic coupling, first introduced in Kessler (1961), occur when a single item or several items of reference is shared by two papers. Taking the data set for this study as an example, two theses sharing five references have five units of coupling between them, which equals one link of a coupling strength of 5. Bibliographically coupled publications are thought to have a cognitive resemblance with each other, something that has been empirically confirmed in several studies (Jarneving, 2006; Kessler, 1963; Vladutz and Cook, 1984; Weinberg, 1974). In the first stage of bibliographic coupling the co-occurrences of the theses reference lists is used to calculate the units of coupling between each pair of theses. In the second stage, following the arguments of Sen and Gan (1983), Glänzel and Czerwon (1995), and Jarneving (2007a), concerning the importance for normalizing for the length of reference lists, the Normalized Coupling Strength (NCS) is calculated for each theses pair. It is defined as:

where r_jk is the units of coupling between document j and document k, and n_j and n_k is the number of references in the reference lists of document j and document k. NCS_j,k is the normalized coupling strength between document j and document k and is a value on an interval between 0 and 1, where the maximum value of 1 is given in the rare case of two document pairs having identical reference lists.

The third stage of bibliographic coupling involved setting a threshold level for the coupling strength and normalized coupling strength to filter away noise in the data (i.e. theses without strong ties or ties without enough strength), and keep only those theses that have stronger ties for the cluster analysis. In order to set this threshold, the impact of different levels of coupling strength on the composition of the data were evaluated, as advised in Jarneving (2006). This is seen in table 1 below.

For example, we see that for a Coupling strength of 3+ (this means only counting thesis pairs having at least 3 units of coupling between them) there are 664 links (thesis pairs) of this strength, these links have an Average normalized coupling strength of 0,09457256, involve 181 theses and are made up of a total of 2513 units of coupling. A coupling strength of 3 seem to be the optimal choice as it does not filter out too many of the original 210 theses, while the links of this Coupling strength have a significantly higher Average normalized coupling strength than those of the starting level.

The quite severe threshold settings of the Normalized coupling strength suggested by some researchers (see Sen and Gan (1983) or Glänzel and Czerwon (1996)) are, according to Jarneving (2007b) not suitable for all kind of data. For such a broad and diverse field as library and information is, the values presented in the table above is to be expected. In fact, they are quite similar to those presented by Colliander (2007). The threshold was set at a coupling strength of 3 with a normalized coupling strength of 0,06, and applied on the thesis level, removing only the theses that did not have at least one link meeting the threshold settings.

Cluster analysis

In the cluster analysis the clustering techniques of Average Linkage, recommended by Colliander (2007), and Ward Linkage, recommended by Janssens, Glänzel, and Moor (2008), were evaluated as candidates for the partitioning of the theses into clusters representing different subjects. These two were applied separately using the normalized coupling strengths derived from the bibliographic coupling as input values. As suggested by previous research (Everitt, Landau, and Leese, 2001; Janssens et al., 2008) the use of mean Silhouette values, interpreted as ‘a measurement of the overall quality of a clustering solution with a specific number of clusters’ (Janssens et al., 2008, p. 610), guided the choice of clustering technique and the number of clusters. This is illustrated in figure 1 below were the mean Silhouette values (going from -1 to +1, higher is better) is calculated for each cluster solution, starting from 2 clusters and increasing, for both Average linkage and Ward linkage.

Balancing the principle of not fragmenting the theses into too many clusters, against the wish for a high mean Silhouette value the Average linkage technique along with a cut-off value producing 38 separate clusters was chosen for the final partitioning. In the graph above we see that when the number of clusters goes up over this value the gains in mean Silhouette value are quite small for eventually to stagnate around 45. Of the 38 clusters produced, 6 were singleton clusters, and thus were removed from further analysis.

Classifying clusters

The keywords of the theses in each clusters were first measured according to a technique called Term Frequency–Inverse Document Frequency (TF-IDF). This technique quantifies ‘the extent of usefulness of terms in characterizing the document [in this study, cluster] in which they appear’ (Akiko, 2003, p. 48). It is defined, in Colliander (2007, p. 28) as:

where freq(i, j) is the frequency of keyword j in cluster i and n(i) is the number of keywords in cluster i. N is the number of clusters and n_c(j) is the number of clusters where keyword j can be found. After applying the algorithm for each keyword in each cluster, the results were ranked from highest to lowest within each cluster. This provided some explanation about the subject focus of each cluster and, together with manual inspection of the theses abstracts, it was possible to classify all 32 of them. As a final step the clusters showing high subject similarity were grouped into overarching themes. All in all, 11 overarching themes were identified, although 5 clusters could not be assigned to a theme.

Results

Library and information science according to the literature cited in master theses

The Price index for the complete data set (leaving out references of the type Web resource and other because of uncertainties in publication date) is 34%. Figure 2 illustrates the 8867 references found in the 210 theses, as divided according to the different source types.

As illustrated above there are basically six types of references that together forms the scientific basis for the master theses: books; Web resources; journal papers; anthology chapters, other student theses; and newspaper articles. They are called the core types, and those receiving 10% of the references will be explored further in this section.

Just by looking at the percentages in the chart above it is evident that books are an important source of information for students of library and information science; over a third of the references goes to books and it is by far the largest of the core types. As a matter of fact, books gather more references than Web resources, journal papers and anthology chapters (the second, third, and fourth largest of the six core types) combined. In figure 3 it is illustrated how the core types of references are distributed over the theses.

The distribution of core types per thesis confirms the supremacy of the book – both the mean and the median is significantly larger than for other core types. The boxes in the boxplot contain the middle half of the data set (called the interquartile range) for each reference type, and give an indication of the citing pattern of a typical master student in library and information science. Reading this plot, we see that half of the master theses cite between 11 and 20 books, 2 and 9 Web resources, 1 and 6 journal papers, 1 and 5 anthology chapters, 1 and 4 student theses, and 0 and 3 newspaper articles. It should be noted, though, that some theses deviate quite a lot from the typical information usage specified within the boxes. These deviations are marked by small circles (outliers) or stars (extreme outliers), and they are especially prominent in the core types Web resources and newspaper articles (although they are visible for all core types). The core authors, defined as the authors that are cited by the most number of theses, are ranked in table 2 below.

About two thirds of the list is populated with individuals who have written, together or alone, course books on various qualitative methods used in the social sciences. Boréus, Bryman, Bergström, Trost, Kvale, Repstad, Widerberg, Philips and Winther Jørgensen all belong to this group, and for each of them the share of citations going to such books are very high (the lowest value being 92,19%). The influence of these authors is also measured by the fact that 82% of the master theses make references to at least one document written by one of them.

The list of core authors is mainly made up of authors from Sweden or the Nordic countries; only one, Bryman, reside outside these countries. Less than a third of the core authors can be considered dealing with questions specific for library and information science (Hansson, Skot-Hansen, Höglund, Limberg).

Books is, as previously stated, the dominant reference type, as a total of 3424 citations are distributed among 2172 edited anthologies and authored monographs. Some books are however used more than others, which is shown by the skewed distribution of those citations. By dividing each book into one of four groups, depending on how many citations it accrued, this is visualized in Figure 4 below.

Most books, 81,63% to be specific, cited by the master theses are only cited the one time, and together they constitute group 4. These books, together with the books in group 3, receive about 80% of the total citations and concern a wide number of subjects. By studying a random sample containing 10% of the books in these two groups it can be concluded that a little more than two thirds of the titles are part of the academic literature, either as works of original research or as overviews of research produced for educational purposes. The remaining third of the books are published outside of an academic setting and consist of, in the following order: various works of nonfiction (ca. 16%); specific guides and handbooks written by practicing librarians or published by associations such as American Library Association (ca. 8%); and works of fiction (ca. 7%). Group 2, made up of the small portion of books that are cited between 4 and 10 times, is mostly made up of books concerning (in descending order) various qualitative methods, perspectives on various types of libraries, knowledge organization, information seeking, sociology of culture, and cultural policy. A few books related to information retrieval, bibliometrics, information, and information systems were also found in this group. We find the 24 highest cited books, sharing 14,16% of the citations between them, in group 1. Of these core books over half are course books on qualitative methods used in the social sciences, and only a smaller portion are books concerning library and information science; most of them relating to libraries and identity, information literacy, and interactions between information seeking and learning.

Web resources is the second largest reference type, and a total of 1463 references go to a wide variety of webpages. It is hard to see any pattern in this category, but if one aggregates citations to websites it is possible make some observations (a list of the ten websites referenced by most number of theses is shown in Appendix B).

Most of the sites belong to rather big organizations devoted to questions regarding libraries, culture, and education. There should be no surprise to see the website of the University of Borås in the top, as it is widely regarded as the top place for education and research in the field of library and information science in Sweden. It is, of course, also the university where the master theses were written, which is important to keep in mind when interpreting the results. Noticeable in the list is also the dominance of Swedish websites.

Journal papers is the third largest reference type. A total of 982 citations are distributed among 852 papers published in various journals. 779 of these documents are cited one time, and only twenty receive more than two citations. Compared to the core books the core papers are much more related to library and information science; in fact, for nearly all of them, the connection is very clear. Many of these papers focus on concepts, models, theories and (qualitative) methods important for the field, or introduces more general methods within a library and information science context. There are also a few studies focusing on cultural policy, public libraries and Web based user-education. It is worth noting that about a fifth of the master theses completely lacks references to papers published in journals.

About 70% of the citations go to papers published in journals that are classified as focusing on Library and information science. The rest of the citations are spread to journals from a variety of subject areas, the three most common being pedagogy and education, media and communication and literary science.

To find the core journals a Bradford analysis (see Bradford (1934) and Diodato (1994, pp. 16–17)) was undertaken: first, the journals were ranked, in descending order, according to the number of published papers referenced by the master theses; and, second, divided into 3 journal zones each containing an approximately equal share of papers. The number of journals in the 3 zones has the proportion 14:57:240 which is roughly equal to Bradford’s law of 1:n:n^2 with a Bradford multiplier, n, of 4. This is visualized in figure 5, with the addition of a bar indicating the number of citations to journal papers in the zone.

Basically the core journals are those journals that publish the largest number of papers referenced by the master theses. They are found in zone 1 and, reading from the chart, represent 4,50% of the total number of journals, together containing 32,28% of the cited papers and receiving 34,73% of the citations. A ranked list of the core journals is provided in table 3.

With exception to perhaps International Journal of Cultural Policy, all core journals are clearly focused on library and information science. A large majority of the core journals are indexed in Web of Science and are there also categorised as part of the field.

Even if most journals referenced are academic journals, this is not always true. For example, in the list above, Library journal and Scandinavian public library quarterly lack peer-review and are more to be considered trade publications than scholarly journals. Nevertheless, their place on the core list indicate that they are important publications for some of the master students.

Library and information science as represented by clustered master theses

The tables in this section reports the result of the bibliographic coupling and cluster analysis. The column listing the highest ranked keyword is left blank if no high ranking keyword was found. Table 4 below contain all clusters that could be assigned to overarching themes.

Five clusters could not be assigned to an overarching theme. One of them has no common focus while the other deal with Mediating literature and Quality Development, Libraries and patrons with disability, Information retrieval, and Information management. The 32 clusters are to a large majority dominated by library or practice-focused theses that use different qualitative methods, such as interviews, focus groups, discourse analysis and ideology analysis, as the main mode of research technique. There are some exceptions to this though: in cluster 30 and 24 the methods used are quantitative and the focus is information scientific and not connected to libraries; and in theme 1 the methods are qualitative but the focus is on literature rather than libraries. There are also several theses in the largest cluster, cluster 10, that focus on information seeking and information behaviour without making much reference to specific library services.

Discussion and conclusions

In summary Swedish library and information science, conceived as a field of education, have the following characteristics:

• Core types: books are the most frequently cited sources of information, followed by internet resources, journal papers, anthology chapters, student theses, and newspaper articles.
• Core authors: the authors cited by the most number of theses are Nordic scholars who have written course books concerning qualitative methods. Only a third of the core authors specialize in library and information science.
• Core books: the most highly cited books focus mainly on qualitative methods, and only a few of them have a specific library and information science perspective.
• Core Web resources: the Web resources cited by the most number of theses are Swedish websites devoted to libraries, culture and education.
• Core journal papers: the most highly cited papers focus on concepts, models, theories or (qualitative) methods in library and information science.
• Core journals: an overwhelming majority of the journals that publish the largest number of papers referenced by the master theses focus on library and information science.
• The research process behind a majority of the master theses, as indicated by the data, involve taking theory and method from books outside of the subject and combining this with journal papers from within the subject.
• The master theses themselves are mostly concerned with library or practice-related questions and the prevailing research methods are qualitative.

The frequency of Nordic researchers in the above core types can most likely be attributed to the data set only including Swedish master theses. Compared to a professional researcher, a student’s perception of the core literature is probably more connected to the geographic placement of the educational institution where s/he attend. Thus, care should be taken when generalizing the findings beyond the context of Swedish library and information science.

Is library and information science to be considered a soft science or hard science? It has been suggested (see for example Meadows (1998) and Price (1970)) that the ‘harder’ the science, the greater the share of references going to journal papers. That more than a third of the references goes to books, while only a tenth goes to papers in scientific journals, is a clear indication that the citing pattern of master students in the field has more in common with that found in the humanities than those found in the natural sciences. The Price index for the data set shows that only 34% of the references goes to literature not older than 5 years. This can be compared with a study by Price (1970) where classical hard sciences such as physics and biochemistry received a Price index of between 60 and 70%, while the social sciences received an index between 40 and 50%, and the humanities a low 10%. According to the types of literature cited, and the low Price index, Library and information science as represented by master theses most definitely should be considered a soft science. This is also supported by how introductory textbooks (e.g., Chowdhury (2008) and Rubin (2010)) present the discipline (although with some reservations), as well as by the result of the bibliographic coupling. Although library and information science involves some quantitative elements that resemble hard science, like the clusters of Information Retrieval and Bibliometrics, the large majority of theses was clustered into groups where methods like discourse analysis, phenomenology, ideology analysis and policy analysis was the main mode of research technique.

When comparing the themes of the clusters produced by bibliographic coupling with the results of previous research we see both differences and similarities. Of the themes reported in table 4 many can be easily placed within the framework of the ten curricular themes reported by Kajberg and Lørring (2005). The exceptions are theme 1, 2 and 5 which are hard to place anywhere in the framework. There exists a considerable overlap in the themes of the clusters produced by bibliographic coupling and the findings of Ohlsson (2004). However, Ohlsson (2004) reports twice as many theses dealing with knowledge organization (theme 3) indicating that students interest in this area has declined over the years. Also, gender and equality (theme 2) along with new media (theme 7) seem to be newly found interests of the students as they do not figure at all in the results of Ohlsson (2004).

According to Larivière et al. (2012, p. 999) previous bibliometric studies, using data from Web of Science, has grouped library and information science into different subfields: library science, information science, and scientometrics. Often the importance of the last two are emphasized, while the last one is thought of as on the decline. Using master theses as a data source to conceptualize library and information science bibliometrically, produces significantly different results. Only a fraction of the theses can be categorized as scientometrics, a somewhat larger share is information science, and the lion share is library science. In comparisons with the clusters of theses produced it is obvious that earlier bibliometric studies have tended to downplay the importance of library and practice-oriented research.

About the author

Pär Sundling is a doctoral student at the sociological institution at Umeå University, Sweden. He received his Bachelor's degree in library and information science and his Master's degree in library and information science from Umeå University, Sweden. His research focus on the contributions of authors and the valuation of authorship. He can be reached at par.sundling@umu.se