1 Diversity measures and network centralities as indicators of interdisciplinarity: case studies in bionanoscience November 30th, 2006 ISMAEL RAFOLS 1 AND MARTIN MEYER 1,2,3 1 SPRU, University of Sussex, Brighton, United Kingdom 2 Steunpunt O&O Statistieken, Katholieke Universiteit Leuven, Belgium 3 Helsinki University of Technology, Institute of Strategy and International Business, Finland Abstract Mapping and evaluating interdisciplinarity poses major challenges, due to its multidimensional character and its inherent conflict with categorisation methods. Here we develop indicators of interdisciplinarity as cognitive diversity that are not dependent on categorisation. To do so, we integrate in a conceptual framework measures of diversity used in ecology and economics, the measures of similarity predominant in bibliometrics and recently proposed indicators of interdisciplinarity based on social network analysis. We carry out two case studies in bionanoscience which illustrate how these indicators do capture the diversity of research topics engaged by an author or a publication, in contrast to category-based indicators that fail to do so. We suggest that these simple and ready-to-use indicators of cognitive diversity may be of potential importance in comparative studies of emergent scientific and technological fields, where claims of novelty and interdisciplinarity are rife but not always justified. Keywords Interdisciplinary research; nanotechnology; nanoscience; diversity; indicators. Introduction In the policy discourse interdisciplinarity is often perceived as a mark of ‘good’ research: interdisciplinary research is seen as more successful at making breakthroughs and generating more relevant outcomes, be it in terms of innovation for economic growth or for social needs. This belief has led to the design of policies aimed at fostering interdisciplinarity, particularly in those fields, such as biotechnologies or nanotechnologies that are regarded as emerging through technological convergence. However, the concept of interdisciplinarity and its variants (multi-, trans- and cross-) is problematic in many respects, if not plainly controversial (Weingart and Stehr, 2000). In the first place, because, given its polysemous and multidimensional nature (Sanz-Menéndez et al., 2001), there has been no agreement so far concerning its most pertinent indicators, or the appropriateness of categorisation methods based on disciplines (van Raan, 2000; Bordons, 2004). Second, although the etymology of inter-, multi-, trans- and cross-disciplinarity suggests that this is a property of research lying between, beyond or across various disciplines, interdisciplinarity is currently widely (and ambiguously) used to mean research spanning over a variety of areas, whether the areas are academic disciplines, technological fields and/or even industrial sectors. Given this use beyond the disciplinary boundaries, Glaser (2006) suggested that ‘interdisciplinarity’ as employed in science policy is a misnomer: cognitive diversity (in relation to either disciplines, specialties, technologies, industries, stakeholders, research fronts or specialties, etc.) would be a more appropriate label. As a result of the problems highlighted above, policies fostering interdisciplinarity appear to be based more on conventional wisdom rather than on an empirical analysis of research practices. The aim of this investigation is to inform policy-making on the dynamics of emerging fields by providing simple measures that can help capture the intensity of interdisciplinarity in the wider sense of cognitive diversity.
Transcript
1
Diversity measures and network centralities as indicators of interdisciplinarity: case studies in bionanoscience
November 30th, 2006
ISMAEL RAFOLS1 AND MARTIN MEYER1,2,3 1 SPRU, University of Sussex, Brighton, United Kingdom 2 Steunpunt O&O Statistieken, Katholieke Universiteit Leuven, Belgium 3 Helsinki University of Technology, Institute of Strategy and International Business, Finland
Abstract Mapping and evaluating interdisciplinarity poses major challenges, due to its multidimensional character and its inherent conflict with categorisation methods. Here we develop indicators of interdisciplinarity as cognitive diversity that are not dependent on categorisation. To do so, we integrate in a conceptual framework measures of diversity used in ecology and economics, the measures of similarity predominant in bibliometrics and recently proposed indicators of interdisciplinarity based on social network analysis. We carry out two case studies in bionanoscience which illustrate how these indicators do capture the diversity of research topics engaged by an author or a publication, in contrast to category-based indicators that fail to do so. We suggest that these simple and ready-to-use indicators of cognitive diversity may be of potential importance in comparative studies of emergent scientific and technological fields, where claims of novelty and interdisciplinarity are rife but not always justified. Keywords Interdisciplinary research; nanotechnology; nanoscience; diversity; indicators. Introduction In the policy discourse interdisciplinarity is often perceived as a mark of ‘good’ research: interdisciplinary research is seen as more successful at making breakthroughs and generating more relevant outcomes, be it in terms of innovation for economic growth or for social needs. This belief has led to the design of policies aimed at fostering interdisciplinarity, particularly in those fields, such as biotechnologies or nanotechnologies that are regarded as emerging through technological convergence. However, the concept of interdisciplinarity and its variants (multi-, trans- and cross-) is problematic in many respects, if not plainly controversial (Weingart and Stehr, 2000). In the first place, because, given its polysemous and multidimensional nature (Sanz-Menéndez et al., 2001), there has been no agreement so far concerning its most pertinent indicators, or the appropriateness of categorisation methods based on disciplines (van Raan, 2000; Bordons, 2004). Second, although the etymology of inter-, multi-, trans- and cross-disciplinarity suggests that this is a property of research lying between, beyond or across various disciplines, interdisciplinarity is currently widely (and ambiguously) used to mean research spanning over a variety of areas, whether the areas are academic disciplines, technological fields and/or even industrial sectors. Given this use beyond the disciplinary boundaries, Glaser (2006) suggested that ‘interdisciplinarity’ as employed in science policy is a misnomer: cognitive diversity (in relation to either disciplines, specialties, technologies, industries, stakeholders, research fronts or specialties, etc.) would be a more appropriate label. As a result of the problems highlighted above, policies fostering interdisciplinarity appear to be based more on conventional wisdom rather than on an empirical analysis of research practices. The aim of this investigation is to inform policy-making on the dynamics of emerging fields by providing simple measures that can help capture the intensity of interdisciplinarity in the wider sense of cognitive diversity.
2
Diversity is a concept used across a range of scientific fields, from ecology to economics and cultural studies, to refer to three different attributes of a system composed of different elements and/or categories (Stirling, 1998; Purvis and Hector, 2000):
• Variety: number of distinctive categories. • Balance: evenness of the distribution of categories. • Disparity: degree to which the categories examined are different from each other.
Notation: Proportion of elements in category i: pi Distance between categories i and j: dij Similarity between categories i and j: sij Indices: Shannon-Wiener ����i pi ln pi Herfindahl-Hirschmann1 ���� i pi
2 Disparity (or similarity) �ij(i≠j) dij (or �ij(i≠j) sij) Stirling �ij(i≠j) dij pi pj
Table 1. Measures of diversity. Many bibliometric studies of interdisciplinarity have examined the variety and balance of disciplines as indicators of degree of diversity using pre-existing categories (see review by Bordons et al., 2004). Many other studies have used similarity measures between in order to visualise the relative ‘position’ of different scientific disciplines and fields (see review by Noyons, 2004; Boyack et al. 2005). Thus, different bibliometric studies on interdisciplinarity have indeed already looked into the three different aspects of diversity listed above. Other attempts to measure interdisciplinarity include: (i) van den Besselar and Heimeriks (2001), who used factor analysis on similarity measures to discriminate the interdisciplinary elements within a set (journals in their case) as those elements that do not fit into the latent classes represented by eigenvectors; (ii) very recently Leydesdorff (2007) who incorporated measures of centrality in social network analysis. The current investigation builds on previous research on mapping using similarity distances defined by co-occurrence (Noyons 2004) and Leydesdorff’s introduction of social network analysis. Here, we use Stirling’s measure of diversity (see Table 1) (Stirling, 1998) but later simplify it to disparity to avoid imposing categorisation. This measure is a sum over the distances between each pair of categories of a set with a weight proportional to the product of their shares. This form incorporates the various properties inherent in the concept of diversity: variety, balance and disparity. In order to apply this diversity measure, we need two previous requirements which will determine the final measure: (i) a categorisation method for partitioning the set; (ii) a definition of distance between categories. Since our aim is to find a measure of cognitive diversity for a given set of publications, we would need in the first place to assign each publication to a cognitive category (e.g. a discipline or research specialty). However, as outlined above, there are several reasons for avoiding categorisation. In the first place, the only multidisciplinary and most widely used categorisation system is the one provided by ISI, which assigns each journal to one or more disciplines. This may work at the journal level, but is very problematic at the paper level given the heterogeneous contents of many journals. In consequence, our first intention was to conduct cluster analysis in order to create self-organised categories for each bibliometric set, following the approach pursued by Schmidt and collaborators’
1 Herfindahl-Hirschman’s index is equivalent to Simpson’s index, used in biodiversity.
3
(2006). However, clustering exercises result in extremely skewed size distributions (few large clusters, many smalls ones, see Schmidt, 2006, p.9), as one might expect from the scale-free characteristics of citation networks (Barabási and Albert, 1999). Due to these skewed distributions, the clustering approach is only feasible for large data sets and even then it is very sensitive on the cutting threshold set. The alternative we pursue here is to take each paper as its own category, with all the categories having the same weight pi. This approach simplifies the measure of diversity to a measure of disparity or dissimilarity, i.e. of how different the various elements are: ����ij(i≠j) dij, with i,j now being the indices for all the elements of the set. Although this description is mathematically more straightforward, we should like to emphasize that it is not a simplification. On the contrary, we may are argue that it is a more complete description as it adds to the diversity measure between clusters we had, a measure of the diversity within clusters –which was assumed to be zero, but is not negligible in skewed distributions that include macro-clusters. As it has been explained in many clustering exercises (e.g. Klavans and Boyack, 2005, p.354) in order to measure the distances or the similarities between a given set of elements we need to make some choices. First, we must select a context (the e.g. a wider set of papers), that will provide the properties or attributes used to measure the distances. Second, we must choose which attributes will be used to compare the units and third we have to adopt a functional form for the similarity or dissimilarity matrix. Network centralities as measures of diversity Since in this study we are interested in mapping the breadth of knowledge sources a set of papers (e.g. publications by an author during a period), we select the co-occurrences of references as the attribute for comparison (bibliographic coupling). This choice implies that we select the references included in our initial set as the context of comparison. Then, we compute the similarity among all the papers of the context using as functional form the Salton’s cosine of the bibliographic coupling (sij = number of references shared by papers i and j divided by the geometric mean of references in i and j). Finally, the distance between two papers is estimated using centrality measures of social network analysis (for an explanation of centrality measures in terms of bibliometrics, we refer to the recent article by Leydesdorff (2007). In a first instance, we use directly the Salton’s cosine value of bibliographic coupling as the similarity between two papers, which a numbers of studies have claimed to be preferable to others (e.g. see a detailed comparison in Boyack et al. 2005). The similarity measure over the whole network is then the sum of the similarity matrix, which is also the sum of the valued degree centrality for all vertices (Sdegree)2. Sdegree = (N*(N-1))-1 ����ij(i≠j) sij, with sij = Salton’s cosine of reference vectors This measure has a notable downside: if two papers do not share any reference, their similarity is set to 0, irrespectively of their location in the wider context provided. In other words: if they don’t share a reference, the distance between a paper in biophysics and one in biochemistry is assumed to be the same as between one in biophysics and one in sociology. In order to take a wider context into account, we propose a second measure based on the length of the geodesic (the shortest path). Here the length of the geodesic dij is the minimum number of links or edges that are needed to connect papers i and j within the network of papers. Two papers are considered to be connected when their similarity value sij exceed a chosen threshold smin
3 . The diversity-dissimilarity measure is then the sum over all the length of geodesics for the connected
2 The measure gives us the network similarity instead of the diversity, because we sum over element-to-element similarities instead of distances. 3 Notice that this procedure converts the sij matrix into a binary matrix, thus losing information about element-to-element distances.
4
papers. After normalisation this yields the average distance between vertices (papers) in the network. Given that this is approximately the inverse of the mean over all the vertices of the closeness centrality, we will choose this well-established index of social network analysis as the second indicator of similarity (Sclose)4: Sclose = (N*(N-1))-1
����ij (i≠j) 1/dij with dij = Number of edges crossed to connect i and j Here we is set smin= 0.05= 1/20, so that the measure can retain the connection between two short research papers with only 20 references each that share one (a typical case in Nature, Science letters), but at the same time dismiss as irrelevant a small number of shared references between review papers (typically with long reference lists). The normalisation of these indices of similarity/diversity to account for size effects is a complex issue that we aim to address in a future investigation. For the purposes of this study, we have normalised the indices so that they take a value in the [0,1] interval and size does not have an effect on the minimum and maximum values.
Research tradition: Ecology � Bibliometrics � Social Network Analysis
Main concept: Diversity � Similarity � Centrality measures
Valued Degree Centrality ����j sij
Main formulations: ����ij(i≠j) dij pi pj
����ij(i≠j) sij where sij=1/dij is Salton’s cosine Closeness Centrality
����j (N-1)/dij
Table 2. Schematic representation of the path followed to construct the theoretical framework used in this investigation.
Here we present two case studies to test these measures against other measures of diversity used in previous studies. In particular we compare Sdegree and Sclose, with measures of diversity based on the number of categories, first using the journals that appear in the references (Njou=number of journals) and second the widely used ISI subject categories (Ncat=number of categories). For these two categorisation methods, we compute the numerical richness (Ncat/N), a normalised Shannon-Wiener index 5 and the Herfindahl-Hirschmann index. When feasible, we also compute the betweenness centrality, which was proposed on empirical grounds by Leydesdorff (2007). The two studies take the different units of analysis within the emerging field of bionanotechnology: first we compare single papers; second, the publication record of a researcher over various periods. We downloaded sets of papers from searches of ISI Web of Knowledge, computed the Salton’s cosine of similarities with our own software and visualised results and computed centrality measures using Pajek (Batagelj and Mvar, 2006), Bibexcel (Persson, 2006) and the statistical packet R (2006) (all freeware). Case study 1: Diversity of single articles on molecular motors In this case, we are interested in tracking the interdisciplinarity of single contributions in molecular motors research. We build on a previous paper investigation that carried out detailed cases studies on interdisciplinary practices in five research projects. It emerged from interviews that while all the cases used techniques and concepts from a variety of disciplines (they were similarly interdisciplinary in this respect), in some cases the project was the continuation of a well-established research tradition, while in others the project brought together different research traditions (the later are more interdisciplinary in terms of knowledge integration). The analysis here shows that while the new measures of diversity 4Closeness centrality for each vertex is defined as Ci= (N-1)/����j dij, with dij being the length of the geodesic between i and j. By summing over i and dividing by N, we obtain the mean of closeness centrality, Sclose. 5 The Shannon index presented here is normalised with the logarithm of the number of papers in a set. This normalized index reflects the evenness or balance of the distribution.
5
Sdegree and Sclose can track this difference in knowledge integration, category-based measures Shannon and Herfindahl fail to do so.
Table 3. Measures of diversity in the reference set of publications. Legend: N: Number of references in paper. Betw: Betweenness centrality. Shan : Shannon diversity. Herf : Herfindahl-Hirschmann diversity measure. Njou: Number or journals in references. Ncat: Number of ISI categories in references. Sdegree and Sclose are the similarity measures as defined above. The form of the indices in the table is presented so that it increases its value with increasing diversity. The two papers highlighted are those presenting a higher diversity according to the centrality measures. Papers within the same box belong to the same project.
0.0
0.5
1.0
1.5
2.0
2.5
0.0 10.0 20.0 30.0 40.0 50.0
1/Sdegree
1/Sc
lose
0.00
0.10
0.20
0.30
0.40
0.0 10.0 20.0 30.0 40.0 50.0
1/Sdegree
Betw
eenn
ess
0.00
4.00
8.00
12.00
16.00
0.0 10.0 20.0 30.0 40.0 50.0
1/Sdegree
1/He
rfind
ahl
Subject Herf
Journal Herf
0.00
0.20
0.40
0.60
0.80
1.00
0.0 10.0 20.0 30.0 40.0 50.0
1/Sdegree
Shan
non
Dive
rsity
Subj. Div.
Journal Div.
Figure 1. Relation between the various diversity indices and 1/Sdegree.
6
We assess the diversity of knowledge sources of one article by using its reference as the set of papers for comparison. This implies making a two step back: take an article, look the references, compute the similarities using the references of the references. The threshold for the computation of Sclose was set at smin=0.05 for all cases except Noji 1997, where in order to avoid the network splitting into two, it was lowered to 0.025. The results are presented in Table 3 and in Figure 1. While we see an agreement between those measures of diversity based on reference similarity, there is no correlation between these measures and those based on journal and subject categorisation. In order to investigate this disagreement, we have plotted the network of references (see Figure 2 through Figure 5). The measures of diversity Sdegree and Sclose can be understood by looking at the number and structure of the clusters formed through similarity links: the two clusters in Noji 1997’s mean higher diversity than the only cluster in Ishijima 1998’s. However, disciplinary categories are not correlated with the lustering: biochemistry and biophysics are equally found in the two clusters in Noji 1997 or in the one cluster in Ishijima 1998. As a result, diversity measures based on ISI subject category do not correlate with number of clusters.
Biochemistry Biophysics Cell Biology
References in Noji 1997
Figure 2. Bibliographic coupling of the reference set in Noji 1997. The graph illustrates that this seminal paper was based on contribution from two research communities. On the right hand side, the researchers working on linear molecular motors (myosin and kinesin). On the left, those working on bio-energetics of rotary motors (F1-ATPase). The figures at the bottom show the papers classified in different disciplines according to ISI subject classification. Both the left and the right clusters have papers from different disciplines.
7
Second, we have also compared Sdegree and Sclose with betweenness centrality. Here we measured the betweenness centrality by placing the examined paper among the network formed by its own references. Figure 1 shows that these measures are only roughly correlated. Figure 5 compares two cases with high diversity in terms of 1/Sdegree and 1/Sclose but different betweenness: we see that betweenness is very sensitive to the overall structure of the network and is an indicator as linker between clusters of the role of the examined paper. If, as in the case of Sakakibara 1999 (right), the reference set is diverse but there are other links between groups already, then betweeness has a much lower value.
Biochemistry Biophysics Cell Biology
Figure 3. Set of references in the publication Ishijima 1998. This is a case of research related to various disciplines but belonging to a well-established line of research. Therefore, the vast majority of its references are related to each other, but they form a cluster where the incumbent disciplines are mixed (most of not spotted nodes fall into ‘Multidisciplinary’ journals). Only for visualisation purposes, smin=0.1.
8
Biochemistry Cell biologyBiochem. Res. Methods
Figure 4. Set of references in the publication Sakakibara 1999. This set integrates literature from the single molecule detection and manipulation based on kinesin and myosin (left cluster), with studies on dynein (right) and dynein role in the axoneme (top). Rather that clear-cut clusters, the network has three or four fuzzy groups, what explains its diversity. Again, these groups do not coincide with established categories.
Figure 5. Betweenness centrality for Noji 1997 (left) and Sakakibara 1999 (right). For each examined article (black spot in the centre), we have measured its betweenness centrality within the network defined by its own references (coloured spot) (smin=0.05).The areas of the node reflect the value of their betweenness centrality. This measure takes a very high value when the article is the only link between clusters (see left: Betw=0.35), but it quickly decreases if there are other links (see right: Betw=0.11).
9
Case study 2: Diversity in the publication set of an author In a second case studied, we aim to trace how diverse is the career of a given researcher. Here we traced the publications of TQP Uyeda, a researcher in molecular motors, from 1983 to 2006 and computed Sdegree, Sclose and Shannon and Herfindahl’s diversities at different periods of his careers6 (see Table 3 and Figure 6). In each of the first three periods, Uyeda’s research was linked through a common topic, which changed as he moved to new labs. However since 2000, he kept the focus he had since the mid-90’s (on the role of Myosin, a molecular motor, in cytokinesis), developed a related topic (conformational changes in myosin) and started a new line of research in the bioengineering side (applications of molecular protein on nanofabricated devices). The indices Sdegree and Sclose are able to track this increase in diversity in the last period, whereas Shannon and Herfindahl’s can’t because the categories used (journals and subject category) cannot always make the distinction between the clusters. The relatively high values of 1/Sdegree and 1/Sclose are attributable to the fact that 3 out of 12 publications were completely unrelated to the others. Comparison of Figure 6 and Figure 7 shows that there is a correlation between research topics and disciplinary categories: in two clusters cell biology is predominant, while in other two we observe a combination of biophysics and biochemistry. Discussion and conclusions In this paper we have presented a conceptual development which we think helps to understand the relation between measures of diversity used in ecology and economics (Shannon, Herfindahl and Stirling’s; see Stirling, 1998), the measures of similarity predominant in bibliometrics (van Raan, 2000; Klavans and Boyack, 2005) and recently proposed measures of interdisciplinarity based on social network analysis (Leydesdorff, 2007). Building on these previous contributions, we have used network centrality measures Sdegree and Sclose as indicators of cognitive diversity for two case studies in bionanoscience. The case studies illustrate that the proposed Sdegree and Sclose do capture the variety of research topics in the references of a publication or pursued by an author, whereas category-based indicators fail to do so. This result is consistent with the conceptual insights on the dynamics of science in the 1970’s (Mulkay, 1974; Small, 1977) which showed that science develops more in terms of ever-changing research fronts and specialties than as a structure of well established disciplines. As a consequence indicators based on categories cannot adequately map emergent fields. Several aspects of the proposed indicators need further exploration. First, their scalability and field dependence need to be investigated in order to make meaningful comparisons between sets of different size or field. Second, the current method to construct the indicators captures the difference between publication of related research, but cannot make the distinction between: (i) elements that are relatively close but not directly related (e.g. two papers in different research specialties within the big area of biochemistry) and (ii) elements that very far apart (e.g. papers on protein conformational changes and Adorno’s social thought). In order to measure this long-range distances, it would be necessary to extend the context for their computation to science-wide maps of the type devised by Klavans et al. (2005). We suggest that the simple and ready-to-use indicators of cognitive diversity presented here may be of special use in comparative studies for evaluation and mapping of emergent fields, were claims of novelty and interdisciplinarity are rife and not always justified. In the policy making process of emerging fields, which by definition are out-of-discipline/field and attract special funding, indicators of interdisciplinarity such as the ones we have developed will be potentially important in order to
6 Since in this instance the object of study (the set of publications) is also the context, we could not think of a simple and meaningful procedure to compute, for a set, betweenness centrality.
10
make the distinction between established developments, conceivable future visions, and the politics and/or rhetoric of technological promise. Acknowledgements This research is supported by an EU postdoctoral Marie-Curie Fellowship to IR and the Daiwa Anglo-Japanese Foundation. We have benefited from comments by J . Gläser and various of our colleagues at SPRU, in particular B. Martin, A. Stirling and S. Katz. The software was developed by Jonathon Read.
Table 4. Measures of diversity in the publications of an author (TQP Uyeda). Legend: see Table 3.
01-06
Pajek
All: 1983-2006
96-00 91-95 83-90
A
A
C
B
E
E D
B
C
C
D
Figure 6. Bibliographic coupling of the publications of TQP Uyeda (1983-2006). Islands were created using the island partition algorithm in Pajek (min-size =5, max-size= 10). In the periods 83-90, 91-95 and 96-00 the researcher focused mainly on one topic . However, since the late 1990s he starts branching into two topics of molecular motors (C: role of Myosin II in cytokinesis; D: conformational change in myosin) and later in 2000’s into investigations of bioengineering applications of molecular motors (E).
11
Biochemistry& related
Biophysics& related
Cell biology Plant biology
Figure 7. Disciplinary ascription of TQP Uyeda publications according to ISI subject categories.
References Batagelj, V. and Mrvar, A. (2006) Pajek. Program forLarge Network Analysi. http://vlado.fmf.uni-lj.si/pub/networks/pajek/
Accessed 13-11-2006. Bordons, M., Morillo, F. and Gómez, I. (2004) Analysis of cross-disciplinary research through bibliometric tools. In Moed,
H.F., Glänzel, W. and Schmoch, U. (Eds.) Handbook of quantitative science and technology research. Kluwer, Dordrecht, 437-456.
Boyack, K.W., Klavans, R. and Börner, K. (2005) Mapping the backbone of science. Scientometrics 64(3), 351-374. Glaser (2006) personal communication. Mulkay, M. (1974) Conceptual displacement and migration in science: a prefatory paper. Science Studies 4, 205-234. Leydesdorff, L. (2007) "Betweenness Centrality" as an Indicator of the "Interdisciplinarity" of Scientific Journals, Journal of
the American Society for Information Science and Technology (forthcoming). Noyons, C. M. (2004) Science maps within a science policy context. In Moed, H. F. et al (eds), Handbook of Quantitative
Science and Technology Research, Kluwer, Dordrecht, 237-255. Purvis, A. and Hector, A. (2000) Getting the measure of biodiversity. Nature 405, 212-219. R Project for Statistical Computing (2006) http://www.r-project.org/. Accessed 13-11-2006. Persson, O. (2006) Bibexcel. A tool-box programme for bibliometric analysis. http://www.umu.se/inforsk/Bibexcel/
Accessed 13-11-2006. Sanz-Menéndez, L., Bordons and M. Zulueta, M.A. (2001) Interdisciplinarity as a multidimensional concept: its measure in
three different research areas. Research Evaluation 10(1), 47-58. Schmidt, M., Glaser, J., Havemann, F. and Michael, H. (2006) A Methodological Study for Measuring the Diversity of
Science. In Proceedings International Workshop on Webometrics, Informetrics and Scientometrics & Seventh COLLNET Meeting, Nancy (France).
Small, H.G. (1977) A co-citation model of a scientific specialty: a longitudinal study of collagen research. Social studies of science 7(2), 139-166.
Stirling, A. (1998) On the economics and analysis of diversity. SPRU Electronic Working Paper. http://www.sussex.ac.uk/Units/spru/publications/imprint/sewps/sewp28/sewp28.pdf Accessed 01-04-2006.
Van den Besselaar, P., and Heimeriks, G. (2001). Disciplinary, Multidisciplinary, Interdisciplinary: Concepts and Indicators. In Proceedings of the 8th
International Conference on Scientometrics and Informetrics - ISSI2001, M. Davis & C. S.
Wilson (Eds.). Sydney: University of New South-Wales, pp. 705-716. van Raan, A.F.J (2000) The interdisciplinary nature of science: theoretical framework and bibliometric-empirical approach.
In Weingart, P. and Stehr, N. (Eds), Practising Interdisciplinarity, University of Toronto Press, Toronto, pp. 66-78. Weingart, P. and Stehr, N. (Eds). (2000) Practising interdisciplinarity. University of Toronto Press, Toronto.
11
Mapping interdisciplinarity in bionanoscienceMapping interdisciplinarity in bionanoscience
Ismael RafolsIsmael Rafols and Martin Meyer, SPRU and Martin Meyer, SPRU University of Sussex, Brighton University of Sussex, Brighton
Strasbourg, March 26Strasbourg, March 26thth 2007 2007
22
Mapping interdisciplinarity in bionanoscienceMapping interdisciplinarity in bionanoscience
1.1. IntroductionIntroduction•• Structure in the New Modes of Knowledge ProductionStructure in the New Modes of Knowledge Production
2.2. NanobiotechNanobiotech as an ID field as an ID field3.3. Research questionsResearch questions4.4. Assessing knowledge integration through diversityAssessing knowledge integration through diversity5.5. Measuring diversityMeasuring diversity6.6. Preliminary ResultsPreliminary Results7.7. Tentative conclusionsTentative conclusions
33
1. Structure in the new modes1. Structure in the new modesof knowledge productionof knowledge production
44
Mapping science technology and innovationMapping science technology and innovation
A caricature of the postwar model (V. Bush, Merton):Normative (science policy) map with clear boundaries
ComputersComputersSemi-conductorsSemi-conductorsSolid state physicsSolid state physics
Empirical studies: a more complex pictureEmpirical studies: a more complex pictureSciSci. Tech and Innovation Studies. Tech and Innovation Studies ( (1970s and 1980s)1970s and 1980s)
• Multiple and complex feed-back processes in the innovation chain.• Technologies are interdisciplinary, products multi-technology.• Boundaries are NOT clear-cut, but diffuse YET critical:
• Interfaces crucial for knowledge trasnf. between Academia-Industry andIndustry-Consumer .• Boundaries are dynamic, contested and socially shaped (Gyerin, 1983, Starrand Giesemer, 1989).• Determinant in the emergence of new disciplines and technologies.
66
New modes of knowledge production (1990s)New modes of knowledge production (1990s)Normative framework for managing the Normative framework for managing the ‘‘knowledge societyknowledge society’’::
•• Mode 2Mode 2 (Gibbons et al., 1994) (Gibbons et al., 1994)•• Triple HelixTriple Helix (Etzkowitz and Leydesdorff, 1996) (Etzkowitz and Leydesdorff, 1996)
Moving from disciplinary science to Moving from disciplinary science to TRANSdisciplinaryTRANSdisciplinary RESEARCH ( RESEARCH (scisci and andtech.):tech.):
•• Interdisciplinary (various scientific domains)Interdisciplinary (various scientific domains)•• Transdisciplinary (various stakeholders)Transdisciplinary (various stakeholders)•• Closely linked to societal needsClosely linked to societal needs•• Temporal and flexible teamsTemporal and flexible teams
CritiquesCritiques•• Is this a system without differentiation? (Shinn, 2002)Is this a system without differentiation? (Shinn, 2002)•• Contrary to systemic efficiency Contrary to systemic efficiency –– benefits from knowledge specialisation benefits from knowledge specialisation
((PavittPavitt, 1998), 1998)
77
Differentiation/Specialisation vs. Integration/ConnectivityDifferentiation/Specialisation vs. Integration/Connectivity
•• Integration makes thisIntegration makes thisefficiency to pay off atefficiency to pay off ata larger scalea larger scale
SYSTEMIC CHANGESYSTEMIC CHANGEtowards a moretowards a moreINTEGRATED systemINTEGRATED system
Means newMeans newdifferentateddifferentatedstructures at thestructures at theprevious boundariesprevious boundaries
(from Stirling, 1998)
More integrated Less integrated
88
2. Nanobiotechnology as an2. Nanobiotechnology as aninterdisciplinary fieldinterdisciplinary field
99
Nanotechnology as a Mode 2 paradigmNanotechnology as a Mode 2 paradigm
Emerging science-based technologies (biotech, nanotech) are Emerging science-based technologies (biotech, nanotech) are viewedviewedas a result ofas a result of
In consequence:In consequence: Specific funding to promote inter- and Specific funding to promote inter- and transdisciplinaritytransdisciplinarity
(In particular research collaborations, industrial and social links)(In particular research collaborations, industrial and social links)
Here, we focus on:Here, we focus on: Interdisciplinarity in academic contextInterdisciplinarity in academic context Bio-nano science/technologyBio-nano science/technology
1010
Nanobiotechnology as an ID fieldNanobiotechnology as an ID field
•• OECD:OECD: ““sits at the sits at the interfaceinterface between physics, biology, chemistry between physics, biology, chemistryand engineering sciences.and engineering sciences.””
•• BBSRC:BBSRC: ““multidisciplinarymultidisciplinary area that sits at the interface between area that sits at the interface betweenengineering and the biological and physical sciences.engineering and the biological and physical sciences.””
•• In the US: In the US: ““NBICNBIC””converging nano-bio-information-cognitive technologies to explain the mindand human behavior by understanding their ‘physico-chemical-biological processesat the nanoscale (…) Unification of science based on unity in nature and itsholistic investigation will lead to technological convergence and a moreefficient social structure for reaching human goals” (Roco & Bainbridge, 2003, )
1111
Difference betweenDifference betweenVISIONVISION (normative) and current practices?? (normative) and current practices??
Empirical studies on nanotech: contradictory evidenceEmpirical studies on nanotech: contradictory evidence
•• Meyer and Persson (1998) suggestMeyer and Persson (1998) suggestthat as a whole the field is morethat as a whole the field is moreinterdisciplinary than interdisciplinary than ‘‘normalnormalsciencescience’’..
• Schummer (2004) found that eachnano-journal publishes articles ofmainly one discipline.
Know. Integration (inter)
Know. Aggregation (multi-)
1212
3. Research questions3. Research questions
1313
Mapping processes of knowledge integrationMapping processes of knowledge integration
How is How is bionanobionano interdisciplinary? interdisciplinary?How is knowledge integration achieved?How is knowledge integration achieved?
‘‘Try to distinguish facts from fads,Try to distinguish facts from fads,perceptions from practicesperceptions from practices’’ PavittPavitt, 2004., 2004.
TEST and CHALLENGE:TEST and CHALLENGE:•• Is really Is really bionanobionano research the result of merging of fields? research the result of merging of fields?•• Are Are bionanobionano groups carrying out ID research or integrating groups carrying out ID research or integrating
knowledge?knowledge?
1414
Mapping processes of knowledge integrationMapping processes of knowledge integrationHow is How is bionanobionano interdisciplinary? interdisciplinary?
How is knowledge integration achieved?How is knowledge integration achieved?
Level of analysis:Level of analysis:•• MICRO: Research group (the academic lab)MICRO: Research group (the academic lab)
strategies for knowledge acquisitionstrategies for knowledge acquisition
•• MESO: Research specialty (knowledge domain)MESO: Research specialty (knowledge domain)dynamics of specialtydynamics of specialty
Methods:Methods:•• Qualitative comparative study (Interviews)Qualitative comparative study (Interviews)•• Bibliometric analysis Bibliometric analysis TODAY TODAY
Policy relevance:Policy relevance:•• What type of funding is appropriate to foster What type of funding is appropriate to foster bionanobionano??
•• ID Collaborations?ID Collaborations?•• ID teams?ID teams?•• ID ID postgradpostgrad schools? schools?
Triangulation forvalidation
1515
4. Assessing knowledge4. Assessing knowledgeintegration through diversityintegration through diversity
1616
How to assess knowledge integration?How to assess knowledge integration?
Focus on a specific field of Focus on a specific field of bionanobionano•• Nanotech is an umbrella and contested term encompassing veryNanotech is an umbrella and contested term encompassing very
heterogeneous fields.heterogeneous fields.•• Generalisation cannot be assumed.Generalisation cannot be assumed.
How can we measure knowledge integration?How can we measure knowledge integration?
Standard method in academic : Standard method in academic : Interdisciplinarity of Interdisciplinarity of publicationspublicationsProblems:Problems:•• Lack of consensus concerning the adequate indicatorsLack of consensus concerning the adequate indicators
(Porter and Chubin, 1985; Bordons et al. 2004; van Raan, 2000)(Porter and Chubin, 1985; Bordons et al. 2004; van Raan, 2000)•• Disciplinary categories are not reliable (ISSI subject classification)Disciplinary categories are not reliable (ISSI subject classification)
total clustering exercise finds 50% of papers outside category cluster.total clustering exercise finds 50% of papers outside category cluster.
Method in firmsMethod in firms’’ technological base: technological base:•• Use relatedness (Use relatedness (BreschiBreschi et al. 2003) et al. 2003)
1717
Knowledge integrationKnowledge integration
Knowledge integration:Knowledge integration:‘‘Bringing together Bringing together previouslypreviously unrelatedunrelated bodies of knowledge bodies of knowledge’’
DynamicsDynamics change changeKnowledge DiversityKnowledge Diversity various knowledge bits various knowledge bits
UnrelatedBodies of knowledge
Relatedknowledge
INTEGRATION(transformation)
HIGH DIVERSITY LOWER DIVERSITY
1818
A definition of diversityA definition of diversity
Missing last week: Diversification without a diversity definition!!Missing last week: Diversification without a diversity definition!!Laursen&SalterLaursen&Salter (2006): (2006): numbernumber of types of partners of types of partnersFai&vonFai&von Tunzelmann (2001): distribution (%) of patents in technologies Tunzelmann (2001): distribution (%) of patents in technologies
Diversity:Diversity:‘‘attribute of a system whose elements may be apportioned intoattribute of a system whose elements may be apportioned into
Characteristics:Characteristics:Variety:Variety: Number of distinctive categories Number of distinctive categoriesBalance:Balance: Evenness of the distribution Evenness of the distributionDisparity:Disparity: Degree to which the categories are different Degree to which the categories are different
1919
Relationship between diversityRelationship between diversityand variety, balance, disparityand variety, balance, disparity
(from Stirling, 1998)
2020
5. Measuring diversity5. Measuring diversity
2121
Measures of diversity (for a set)Measures of diversity (for a set)
Notation:Notation: ppii Proportion of elements in category Proportion of elements in category iiddijij Distance between categories Distance between categories ii and and jj
Indices:Indices:Shannon-WienerShannon-Wiener (entropy)(entropy) ∑∑i i ppii lnln p pii
HerfindahlHerfindahl (concentration) (concentration) ∑∑ i i ppii22
Attention: Most indices do not include distanceAttention: Most indices do not include distance
2222
Diversity/similarity (for a set)Diversity/similarity (for a set)
StirlingStirling ∑∑ij(iij(i≠≠jj) ) ppii ppjj ddijijtakes into account variety, balance and disparity.takes into account variety, balance and disparity.
Here we use bibliometric sets (i.e. publications):Here we use bibliometric sets (i.e. publications):•• Easy to define similarities/distancesEasy to define similarities/distances•• Very difficult to define categories Very difficult to define categories What is a body of knowledge? What is a body of knowledge?
There are clusters not defined by discipline (e.g. There are clusters not defined by discipline (e.g. Soc.Net.AnalSoc.Net.Anal.).)
Solution: use each element (publication) as its own categorySolution: use each element (publication) as its own category
Where Where ssijij is the similarity between categories i and j is the similarity between categories i and j
2323
Measure of similarity/relatednessMeasure of similarity/relatedness(between elements)(between elements)
Here we use as similarity measure:Here we use as similarity measure:
•• ssijij = (# refs. = (# refs. i-ji-j share)/ share)/sqr(#refsqr(#ref i i *#ref *#ref jj))SaltonSalton’’ss cosine of bibliographic coupling cosine of bibliographic coupling
•• ssijij = 1/(shortest path between = 1/(shortest path between ii and and j)j)where publication i and j are connected if they share awhere publication i and j are connected if they share apaperpaper
Other common measures:Other common measures:•• Co-citationCo-citation•• Co-occurrences of technological fields (IPC)Co-occurrences of technological fields (IPC)((BreschiBreschi, , LissoniLissoni and and MalerbaMalerba, 2003), 2003)
2424
S
Similarity Measure 1:% of sharedreferences = 0.06
Similarity Measure 2:1/(min distance)= 2.0
2525
Similarity measures in a Similarity measures in a SNASNA’’ss form form
SSdegreedegree = = ∑∑ij(iij(i≠≠jj) ) ssijij = = Sum over Valued Sum over Valued Degree CentralityDegree Centralitytells about tells about short rangeshort range interactions interactions
SScloseclose = = ∑∑ijij 1/d1/dijij = = Sum over Sum over Closeness CentralityCloseness Centralitytells about tells about long rangelong range interactions interactions
Derivation path of diversity measureDerivation path of diversity measure
What is worth going for a excursion into diversity???
2727
6. Preliminary results6. Preliminary results
2828
Preliminary resultsPreliminary results
•• What is the cognitive diversity of one publication?What is the cognitive diversity of one publication?
For a set of publications on For a set of publications on biomolecularbiomolecular motors motors
take the reference set of each publication, measuretake the reference set of each publication, measureits diversityits diversity
compare with standard diversity measures usingcompare with standard diversity measures usingISSI subject classificationISSI subject classification
validate the results using a previous interview tovalidate the results using a previous interview toauthors regarding this particular research effortauthors regarding this particular research effort
2929
Biochemistry Biophysics Cell Biology
Joining two research specialtiesinterdisciplinarity within each specialty
Bibliographic coupling ofreferences in Noji 1997
1/S1/Sdegree degree = 39= 39
3030
Biochemistry Biophysics Cell Biology
Interdisciplinaritywithin ONE specialty
Bibliographic coupling ofreferences in Ishijima 1998
1/S1/Sdegree degree = 24 = 24
3131
Biochemistry Cell biologyBiochem. Res. Methods
Bibliographic coupling ofreferences in Sakakibara99
Diversity of one publicationDiversity of one publication
3333
Diversity of one researcherDiversity of one researcher
•• What is the cognitive diversity of one researcherWhat is the cognitive diversity of one researcher’’sspublications?publications?
•• How does it evolve?How does it evolve?
take his/her publication set, measure diversity in take his/her publication set, measure diversity indifferent periodsdifferent periods
Caveat: this measure can be affected by normalisation
3434
01-06All: 1983-2006
96-00 91-95 83-90
A
A
C
B
E
E D
B
C
C
D
Diversity in researcher’s pub set
3535
Diversity within journal setDiversity within journal set
Lab-on-a-Chip, 2005
1/Sd =102
J. Nanotech & Nanoscience, 2005
1/Sd =435
Threshold = 1
3636
6. Tentative conclusions6. Tentative conclusions
3737
Work in progressWork in progress
•• Indicator that can capture cognitive diversityIndicator that can capture cognitive diversityby-passing categorisationby-passing categorisation
•• Diversity (similarity) indicator, simple to construct fromDiversity (similarity) indicator, simple to construct frombibliometric databibliometric data
Next stepsNext steps•• Temporal evolution over research specialtyTemporal evolution over research specialty
E.g. Lab-on-chip articles.E.g. Lab-on-chip articles.•• Validation over larger data set.Validation over larger data set.•• Small world Small world –– local clustering vs. global connectivity local clustering vs. global connectivity
Open issuesOpen issues•• Normalisation by size???Normalisation by size???
3838
AcknowledgementsAcknowledgements
Funded by a Marie CurieFunded by a Marie CuriePostdoctoral FellowshipPostdoctoral Fellowship(2006-08) and(2006-08) and
the Daiwa Anglo-Japanesethe Daiwa Anglo-JapaneseFoundation (2005-07).Foundation (2005-07).
Accepted for the 11Accepted for the 11thth ISSI Conference, June 2007 ISSI Conference, June 2007Preprints available atPreprints available atSee See www.sussex.ac.uk/spru/www.sussex.ac.uk/spru/irafolsirafols
