IJPDLM 28,5

328 Received September 1996 Revised May 1997, March 1998, May 1998

A social network analysis of business logistics and transportation Diane M. Phillips St Joseph’s University, Philadelphia, Pennsylvania, USA and

Jason Keith Phillips Ursinus College, Collegeville, Pennsylvania, USA

International Journal of Physical Distribution & Logistics Management, Vol. 28 No. 5, 1998, pp. 328-348, © MCB University Press, 0960-0035

The evolution of a scientific discipline The origins of marketing thought have been traced to the beginning of this century when business people realized that marketing was more than a mere activity, it was an idea worthy of study (Bartels, 1965). Although most of the research done at that time focused on the movement of farm products to market, a significant advancement occurred when universities first offered courses in “marketing”. This occurred in 1905 at the University of Pennsylvania, in 1909 at the University of Pittsburgh, and in 1910 at the University of Wisconsin (Bartels, 1965). Marketing as a scientific discipline can trace its roots to 1936, or the date at which the first article appeared in The Journal of Marketing. Now, more than 60 years later, the academic study of marketing has been firmly established. The fields of business logistics and transportation, like consumer behavior, retailing, and sales, evolved from this general conception of marketing. In fact, within a few years of the genesis of the marketing concept, Shaw (1916) noted that marketing was composed of two conceptual halves: demand creation and physical supply. Over time, these halves have evolved with demand creation being studied within such subdisciplines as retailing, advertising, and sales, and physical supply evolving into the modern subdisciplines of business logistics and transportation. Fifty years after Shaw’s observation, Bowersox (1966) declared that the fields of business logistics and transportation were in a state of semi-maturity. Since that time, the fields have continued to evolve and mature. This case study describes one way in which business logistics and transportation changed over time. That is, we describe the changing communication patterns that occurred between and among the members of its social network. Previous research on the development of other disciplines (e.g. services marketing) has found that a new academic field can develop because it fulfills a need for academics or practitioners (Berry and Parasuraman, 1993). The authors wish to extend a note of thanks to Martin Kilduff and the two anonymous reviewers for their invaluable comments and suggestions on an earlier draft of this paper.

We start from the same broad hypothesis in this study – that the fields of A social network business logistics and transportation emerged to facilitate information analysis exchange for academicians and to provide a source of new and emerging techniques for practitioners. We argue that the discipline continues to fulfill this function, and the way it fulfills this function has changed over time. The fields continue to grow with respect to their size, interest, and importance. In addition, 329 many scholars who were instrumental in the development of business logistics and transportation are still active in the field today and more and more doctoral students are choosing to initiate their careers in the field. Further, many practicing managers received bachelors or masters degrees in business logistics and transportation. In order to better understand the changing characteristics of these related fields, we have chosen to study an artifact of this process, citations in journals. Admittedly, other historians may choose different means by which to study the process of the maturation and evolution of a scientific discipline. Historians can administer surveys, conduct interviews, or examine documents such as minutes from meetings or other published material (see Berry and Parasuraman, 1993). However, we deemed the analysis of the artifacts of communication between and within journals to be concrete evidence of where knowledge is generated and used. The peer review process ensures that the articles that are published are, to some extent, intellectually stimulating to other researchers. We argue that in their research, scholars read and cite articles that they find interesting, insightful, and that aid them in their thinking. After articles are published, the information is disseminated to other academicians in other disciplines and, importantly, to practitioners who may use some of the knowledge and findings to aid their own business practices. The cycle of knowledge generation and dissemination begins again when researchers study businesses, conduct analyses, develop theories, and publish their findings. Thus, one article builds on the next and the process of knowledge generation and dissemination continues. It is in this way that an academic discipline grows and matures. One sign of maturity in a scientific discipline occurs when that discipline becomes interested in its own patterns of scholarly communication (Borgman, 1990; Everett and Pecotich, 1993). Several scholars have already examined the nature of the fields of business logistics and transportation (Emmelhainz and Stock, 1989; Fawcett et al., 1995; Ferguson, 1983). We wish to further that investigation. The following study accomplishes two specific goals. First, we introduce social network analysis techniques to the business logistics and transportation community as a useful tool with which to study the dynamic flows of communication between members of a social network. Second, we describe a wide variety of techniques and apply them to the business logistics and transportation citation network. In doing so, we track the changing communication patterns across two separate time periods to describe the evolution and maturation of the fields of business logistics and transportation.

IJPDLM 28,5

330

Social networks Social network analysis is a fairly new technique which traces its earliest beginnings to the 1960s (Scott, 1991) and has just recently been introduced to the field of business. A social network can be defined as a group of social actors that interrelate or exchange information with one another. The study of a social network that exchanges information between and among its members provides interesting insights into how knowledge is spread throughout a community. In addition to providing a better understanding of a discipline’s communication patterns, social network analysis can also be used to better understand and analyze networks of interrelated scientific disciplines. As such, social network analysis refers to a body of quantitative and qualitative measures which are used to better understand the relationships between and among members in a given social network (Scott, 1991). The relationships under scrutiny could include friendship, influence, or in the case of a scientific discipline, patterns of communication or strength of association between members in a scientific community. Citation analysis Scientific articles cite and are cited by other scientific articles and an examination of these citation patterns can illuminate important flows of communication as well as other relationships within the network such as prestige or influence. This can be accomplished by this technique’s unique ability to provide information on the identity of journals which make and receive citations (“directional” data), as well as information on the total number of citations those journals make and receive (“valued” or “strength” data). Citation analysis research has, for example, examined a global network of scientific articles (Price, 1965), untangled smaller networks of scientific articles (Noma, 1983) and described the structure of a subdiscipline of a scientific community (Cote et al., 1991; Gatrell, 1984; Leong, 1989). On a micro level, citation analysis is a technique which can provide greater information about the citations made to and received from journals, authors, books, etc. At a more macro level, the overall purpose of citation analysis is to provide greater information about communication patterns within and between scientific disciplines. Social networks and citation analysis working in concert When studying the evolution of a scientific discipline, social network analysis can be combined with citation analysis to help us untangle and better understand the patterns of communication in a given social network. With respect to business logistics and transportation, several studies have attempted to describe the nature of the discipline using survey research (Ferguson, 1983; Emmelhainz and Stock, 1989; Fawcett et al., 1995). To date, no studies have used both social networks and citation analysis together to study these fields. While these techniques may be mathematically cumbersome and timeconsuming, they represent a vast improvement over previous procedures used in the analysis of the exchange of information between and among journals in

a scientific discipline. Together, these techniques have the ability to account for A social network all two-way connections between every member of an academic network by analysis analyzing the sending and receiving patterns of citations of all academic journals within a given scientific network. These techniques can also identify the influential members, or “invisible college”, in a scientific community (Crane, 1972). An invisible college is a tightly knit network of scientists at the center of 331 any particular area of research (Burt, 1982). These techniques are not bound to the use of raw frequency counts in their calculations, which can sometimes mask associations between network members. Finally, these techniques do not make any a priori assumptions about prominent or influential members in the network. They let the data speak for itself (see Figure 1). After an examination of the existing literature on network and citation analysis, it is clearly evident that a void exists with respect to business logistics and transportation. Although the results of a social network analysis provide a Survey Research

Social Network Analysis

Citation Analysis

Examples:

Examples:

Examples:

• Ferguson (1983) – a ranking of publication prestige level • Emmelhainz and Stock (1989) – an estimate of the highest ranking logistics and related journals with respect to their impact on the discipline • Fawcett, et al. (1995) – a rating of academic logistics and transportation journals

• Reingen and Kernan (1986) – an assessment of a network’s prestige, subgroups, and flows of information using word-ofmouth referrals • Iacobucci and Hopkins (1992) – an analysis of power, cooperation, conflict resolution, and the management of expectations using word-of-mouth networks

• Hamelman and Mazze (1973) – an identification of the most “active” journals with respect to two other important journals • Leong (1989), Cote et al. (1991), Zinkhan et al. (1992) – an analysis of communication patterns of journals in consumer research • Hoffman and Holbrook (1993) – an analysis of authors in consumer research

Social Networks & Citation Analysis Working in Concert Advantages: • does not rely on raw frequency counts to compile lists of, for example, frequently cited authors, journals, etc. which can mask linkages and associations between network members (cf., Leong , 1989; Cote et al., 1991; Hoffman and Holbrook, 1993; Zinkhan et al., 1992) • allows for an analysis of the two-way flow of information between network members • allows the data to speak for itself, rather than making an a priori assumption about a particular network member being most prominent (cf., Hamelman and Mazze, 1973; Leong, 1989; Cote et al., 1991) • identifies members in the “invisible college” Examples: • Burt (1982) – an identification of an “invisible college” within sociology • Everett and Pecotich (1993) – a multidimensional scaling of the network • Salancik (1986) – an identification of powerful and influential members in the network as well as an assessment of the span of that influence

Figure 1. Social networks and citation analysis working in concert

IJPDLM 28,5

332

more veridical account of the actual communication patterns in an academic network, as well as other informative statistics, the use of such techniques in business logistics and transportation has been non-existent. To date, no studies have used a comprehensive social network approach to citation analysis to identify subgroups, prominent journals, flows of information, or “invisible colleges” within the academic disciplines of business logistics and transportation. Further, no previous studies have taken a longitudinal approach to understanding how an academic field evolves and matures over time. Accordingly, the purpose of this study is twofold. First, we wish to introduce social network analysis techniques as applied to citation analysis to business logistics and transportation. Second, we want to apply those social network analysis techniques to patterns of academic citations to more clearly understand how in one case, we can better understand how these two related disciplines which share a common history may have changed over time. Business logistics and transportation represents an ideal case study with which to track the changing citation patterns across time. Early monographs describe the discipline as originally evolving out of marketing (Shaw, 1916) and reaching “semi-maturity” in the 1960s (Bowersox, 1966). Now, 30 years later, it would be interesting to see if business logistics and transportation has actually reached maturity. If it has, it will be more likely to draw knowledge and information from itself than from other disciplines (Borgman, 1990; Everett and Pecotich, 1993). In addition, its existence should fulfill a purpose (Berry and Parasuraman, 1993) or perform a role for academicians or practitioners. Method In order to study the evolution of business logistics and transportation, a longitudinal analysis of citation data was performed. Specifically, social network analysis was performed on a network of academic journals across two separate time periods. The two time periods were selected in order to allow a ten-year interim time interval, enough time for communication patterns to shift and change. The two time periods selected were: 1981-1983 and 1991-1993. This ten-year time frame was selected because it represented a period of great interest and activity in the field and because citation analyses often use time periods that are a decade apart (cf. Doreian, 1985; 1988). In addition, the Journal of Business Logistics was first published in 1978 and we believe it would be interesting to assess how the influence of this journal changed over time. Each time period included all data on all citations between all journal pairs within the network. Further, unlike earlier studies, no a priori assumptions were made with regard to the identity of prominent journals. Journal selection We started with the overall goal of including any and all academic journals that may be important to the field. We therefore consulted earlier studies which included lists of journals and extracted a set of journals that, we believe,

reasonably approximates the relevant social network for business logistics and A social network transportation. It encompasses a diverse set of journals that both influence and analysis are influenced by the fields. The journals selected for inclusion in the network represented all spheres of the business logistics and transportation disciplines and included topic areas such as logistics, marketing, transportation, engineering, general business, and management science. 333 The journals which compose the matrix were selected via a five step process. This stage of the research was critical because once the matrix was developed and data collection commenced, adding additional journals would entail starting the data collection stage over. The first stage of journal selection consisted of selecting those journals classified by Emmelhainz and Stock (1989) as being specialty publications directed either toward an academic or a mixed (academic and practitioner) audience. This was used as a starting point because one purpose of the Emmelhainz and Stock (1989) study was to identify important journals in the fields of business logistics and transportation. Second, given that the primary purpose of this study was to study the case in which an academic discipline’s communication patterns change over time, those journals identified by Allen and Vellenga (1987) that contained at least half academic-authored articles were selected. Next, any journals identified by Ferguson (1983) as having a readership level of at least 20 per cent and that were in journal format (e.g. footnotes, endnotes, or references and issued a fixed number of times per year) were added. Fourth, those journals from Ferguson (1983) that Emmelhainz and Stock (1989) identified as being directed primarily toward a practitioner audience were deleted. Finally, any remaining journals that were not refereed publications during both time periods were deleted. The resulting network of journals represents the primary academic-related journals relevant to the business logistics and transportation scientific community (see Table I) [1]. Data collection The data for this study were collected by manually counting citations in each of the 15 journals for each year examined in the study. Manual counting was necessary for two primary reasons. First, a number of the journals within the matrix are not supported by the Social Sciences Citation Index or the Journal Citation Reports, which are often used in compiling data in citation analyses. Second, these sources have been previously criticized as having numerous reliability problems (Hoffman and Holbrook, 1993; Reed, 1995; Rice et al., 1989). These errors include discrepancies between citing and cited data, changed or deleted journal titles, the use of unusual or inconsistent journal abbreviations, and missing or unavailable data (Rice et al., 1989). Consequently, matrices were manually constructed for each year (1981, 1982, 1983, 1991, 1992, 1993) and thus represented the number of times each journal cited the other journals within the matrix as well as the total annual number of citations within each journal.

IJPDLM 28,5 Journal

334

Decision Sciences Industrial Marketing Management International Journal of Physical Distribution & Logistics Management International Journal of Purchasing and Materials Management Journal of Advanced Transportation Journal of Business Logistics Journal of Marketing Journal of Marketing Research Journal of Transport Economics and Policy Logistics and Transportation Review Management Science Naval Research Logistics Transportation Transportation Journal Transportation Quarterly

Journal rankings Ferguson Emmelhainz Fawcett et al. Abbreviation (1983) and Stock (1989) (1995) DS IMM

7 11

* *

* *

IJPDLM

6

3

4

IJPMM JAT JBL JM JMR

9 * 4 3 5a

4 * 1 * *

6 10 2 * *

* 2 5a 8 10 1 *

* * * * * 2 *

3 5 * 7 9 1 8

JTEP LTR MS NRL T TJ TQ

Notes: * = No ranking available a = Tied rankings Table I. Ferguson (1983) was published at the end of period 1, Fawcett et al. (1995) was published just The network of journals after period 2, and Emmelhainz and Stock (1989) was published between the two periods.

Matrix construction For each of the two three-year time periods, the journals were combined into directional, valued 15×15 matrices. The six matrices were then summed to create a period 1 matrix (1981-1983) and a period 2 matrix (1991-1993). Threeyear time periods were used in order to smooth the impact of any unusual variances in a given year while still maintaining the integrity of the citation links. Thus, the three-year totals represent a more accurate picture of typical sending and receiving patterns for the two time periods. The two matrices were then subjected to row normalization in order to control for the fact that some journals were more active than other journals. Row normalization corrects for the fact that some journals have a much larger output of citations (because they publish more articles, because authors in that journal make more citations, etc.). For example, in period 1, MS made 9,826 citations to other journals while JAT made only 727 citations. Data on these frequent citers would completely overwhelm other data in the matrix. Therefore, it was determined that this potential source of error should be controlled by normalizing the data (for a more detailed discussion, see Doreian, 1985, 1988; Phillips et al., 1998). A 16th column was subsequently added to each

matrix which included totals of all citations made by the core network of 15 A social network journals. The rows of the two matrices were then subjected to a marginal analysis normalization using UCINET IV (Borgatti et al., 1992)[2]. By forcing the sum of the elements to be 100, cell entries could be likened to percentages of citations given or received. All further analyses were performed on the directional normalized 15×15 matrices, leaving out the 16th total column (see Appendix).

335

Results Descriptive analyses A quick perusal of the matrices revealed that self citations accounted for most of the citations made in the network. Of all the journals in period 1, JMR was the most active self citer with 14 per cent of the total citations made within the network. In period 2, IJPMM made most of the self-citations, which accounted for 15 per cent of the total citations made in the network. Self-citations are an important consideration in social network analysis because, among other things, they represent a tendency for a given journal to be self-contained. In other words, these journals less frequently obtain knowledge from other journals. Interestingly, the journals which make up the core of the field of business logistics and transportation all became more self-contained in the second period. JBL’s self citations increased from 0.90 to 7.05 per cent, IJPDLM’s self citations increased from 2.67 to 3.06 per cent, and TJ’s self citations increased from 2.64 to 5.84 per cent. We then computed matrix correlations for the row-normalized data between the two time periods. Despite fluctuations in the sending and receiving patterns of specific journals in the two time periods, the overall correlation between the two matrices was 0.85. Thus, the overall citation network was rather stable across time. A closer analysis of the particular differences through the use of measures of centrality, a measure of groupings, multivariate procedures, and position in the network will reveal the finer distinctions and changes that took place across time. Measures of centrality As a way of assessing prominence in a network, measures of centrality help identify prominent actors in a network that are particularly visible relative to other actors in the network (Knoke and Burt, 1983). Specifically, a journal is considered “locally central” if it has a large number of direct citations with other journals in its immediate environment and “globally central” when it has a position of strategic significance in the overall structure of the network (Scott, 1991). Perhaps the easiest and most straightforward way to measure the centrality of a network member is by assessing its degree (Scott, 1991). As such, the first centrality analysis performed was an assessment of degree. More central journals will be those which receive a larger number of citations from other journals (“indegree”) than they make (“outdegree”) (Knoke and Burt, 1983). For both periods, the overall indegree of the network was considerably higher than its outdegree (see Table II). This indicates that the network itself made more citations within the network than it did outside the network. With respect

IJPDLM 28,5

336

Table II. Journal centrality: inDegrees and outDegrees

Journal

InDegree

DS 7.78 IMM 8.09 IJPDLM 4.95 IJPMM 16.28 JAT 1.62 JBL 1.64 JM 21.87 JMR 31.62 JTEP 15.22 LTR 5.74 MS 33.61 NRL 8.92 T 6.37 TJ 5.78 TQ 6.40 Network Centralization 180.4%

Period 1 OutDegree

Difference

InDegree

Period 2 OutDegree

Difference

12.05 19.41 9.67 19.19 5.89 16.46 15.95 18.63 9.20 5.76 12.63 14.05 7.79 4.74 4.47

–4.27 –11.32 –4.72 –2.91 –4.27 –14.82 5.92 12.99 6.02 –0.02 20.98 –5.13 –1.42 1.04 1.93

9.08 11.16 8.67 21.29 2.08 13.28 27.45 27.40 12.05 8.96 33.90 7.84 5.12 13.03 4.02

14.05 18.24 17.07 19.27 4.72 20.25 17.39 20.25 9.38 13.63 10.54 14.62 7.18 13.95 4.79

–4.97 –7.08 –8.40 2.02 –2.64 –6.97 10.06 7.15 2.67 –4.67 23.36 –6.78 –2.06 –0.92 –0.77

63.4%

–

166.6%

54.1%

–

Note: The difference columns provide the most telling evidence of whether or not a journal is a net receiver or sender of information in the network. In period 1, for example, MS is the biggest net sender of information in that many more journals cite MS than are cited by MS.

to the centrality of different journals, MS had the highest positive difference in its degree score for both period 1 (20.98) and period 2 (23.36). This analysis reveals that the logistics and transportation related journals are relatively less central to the network in both periods. Interestingly, this measure indicates that IJPDLM and TJ become less central in the network over time (difference scores: –4.72 to –8.40 and 1.04 to –0.92 respectively) and JBL becomes relatively more central to the network (difference score: –14.82 to –6.97). In order to confirm the aforementioned results related to centrality, the “betweenness” of the journals within the matrix was examined. Developed by Freeman, betweenness measures the extent to which a particular point lies “between” the various other points in the graph (Scott, 1991). Journals which lie between many other journals in the network may play a specific role in channeling information from one part of the network to another. These journals may essentially act as gatekeepers or brokers by controlling the flow of information between other members in the network (Scott, 1991). The relative differences in journal scores are indicative of more or less “betweenness”. The network as a whole underwent a distinct change between periods 1 and 2 (see Table III). In period 1, TJ, JM, and MS were very central to the network in that they were located between many other journals. In period 2, however, very few journals performed the role of gatekeeper. The betweenness scores were much lower overall, and the gatekeepers identified in period 1 were no longer active gatekeepers in period 2. In addition, IJPDLM underwent a remarkable increase

Journal

Period 1 Betweenness score

Period 2 Betweenness score

DS IMM IJPDLM IJPMM JAT JBL JM JMR JTEP LTR MS NRL T TJ TQ

5.52 4.96 2.43 2.90 0.25 3.14 16.58 15.25 1.25 12.41 16.50 0.14 10.76 28.10 0.80

7.38 1.91 12.49 4.84 0.63 7.03 5.78 4.91 4.90 7.12 4.00 1.65 5.33 9.38 10.63

Note: Examining relative differences in betweenness scores within a time period and between time periods is the most useful interpretation of this table. IJPDLM and TQ, for example, become much more “between” in period 2.

in betweenness between the two time periods (from 2.43 to 12.49), leaving IJPDLM as the journal most “between” other journals. In comparison to period 1, period 2 seems to be characterized by more direct communication between journals rather than the channeling of communication between important actors in the network. Within a network of journal relations, each journal contributes citations to and receives citations from other journals in the network. Journals that receive a large number of citations in proportion to the citations they make to other journals are generally thought to be more central to the network. A central location in the network may be indicative of a higher level of prestige. Prestige can be assessed with the help of a directional matrix such as the ones used in this study, which identify both the number of citations made and those received by all pairs of journals. Two measures were used to assess centralization in the network: degree and betweenness. Centrally located journals are those that received proportionately larger indegree scores than outdegree scores and/or had high betweenness scores. The centralization procedures indicated that while several key journals may have played an important central role in the first period, as a whole, the second period had fewer prestigious players and more free flowing, direct exchanges of information. Measure of groupings An analysis of cliques was the next procedure utilized. Cliques require that every member in a given grouping is directly connected to every other member in the grouping (Scott, 1991). As part of the clique procedure, UCINET provides

A social network analysis

337

Table III. Journal centrality: betweenness

IJPDLM 28,5

338

Table IV. Measures of grouping

a hierarchical clustering of the cliques such that progressively more and more journals are added to the most central clique in the network. When comparing the results of the hierarchical clustering in period 1 with period 2, the cliques that emerge are quite different (see Table IV). In period 1, the most central clique is composed of DS, MS, and JBL. The first layer adds IJPDLM, JTEP, LTR and TJ. The second layer adds IMM, and JM. Finally, the third layer adds all the remaining journals in the network. The pattern of cliques in period 2 looks fairly different in that the most central clique contains JMR, LTR, IJPDLM, and JM. The first layer adds JTEP and TJ. The second layer adds MS and T, while the third layer adds DS, JBL, IJPMM, and TQ. The fourth layer adds JAT and NRL, and the fifth layer adds IMM. This analysis indicates that the network of journals is fairly unorganized in period 1 with three layers of journals surrounding a central clique. It was quite surprising to find JBL, a relative newcomer to the network, in a central clique with DS and MS [3]. In this time period, the first layer consists of transportation and logistics journals while the second layer consists of marketing journals. The fact that JM and IMM are categorized in a fairly peripheral clique could be indicative of these two journals being more interdisciplinary and far-reaching in their content than the journals identified in more tightly structured cliques. Not as many journals are directly connected with them. With the exception of DS and MS, the remaining journals in the more centrally located cliques are all transportation and logistics related journals. It makes sense that these journals would bond together to share information and influence at this relatively early stage of maturity. Period 2 contains a far different mix of journals at every level. The most central clique in this period contains JMR, LTR, IJPDLM, and JM. While the first layer can be described as primarily “transportation”, subsequent hierarchical layers add cliques which contain a mixture of management, transportation, and logistics journals. This result again supports the findings of the centrality measures by indicating that the latter network is characterized by more direct communication between member journals. At the heart of the network is a clique that contains both marketing and logistics related journals. Although these journals are very different in content and audience, the analysis of cliques indicates that they communicate directly with one another.

Most central clique Layer 1 Layer 2 Layer 3 Layer 4 Layer 5

Period 1

Period 2

DS, MS, JBL IJPDLM, JTEP, LTR, TJ IMM, JM JMR, T, IJPMM, TQ, JAT, NRL

JMR, LTR, IJPDLM, JM JTEP, TJ MS, T DS, JBL, IJPMM, TQ JAT, NRL IMM

Multivariate procedures A social network The first multivariate procedure performed was a multidimensional scaling analysis (MDS) analysis. This procedure uses a matrix of distances and arranges those points on a two-dimensional space such that the Euclidian distances between the points represent (as closely as possible) the original distances in the input matrix (Borgatti, et al., 1992). Conceptually, the MDS can be interpreted to 339 represent a two-dimensional recreation of the communication patterns between the different network members. Members that are close to one another on the mapping have stronger communication ties. As can be seen in the mapping of period 1, there appears to be a distinct grouping of management and logisticsrelated journals to the left and another grouping of transportation-related journals to the right (see Figure 2). The two marketing-related journals are located on the left, close to the management journals. Two of the primary journals in the field of business logistics and transportation, JBL and TJ, are located close to the center of the entire network. This may indicate that they send and receive citations from many different sources. In this time period, IJPDLM is located with the management and logistics journals. Period 2 shows quite a different account of the location of journals in the network (see Figure 3). Although in general the transportation journals are still to the right and the management and logistics journals are still to the left, this network seems much more tightly collected around the midpoint of the map (with the exception of JAT which appears to be very peripheral to the network in this time period). This network appears much more cohesive and lacks the distinct break between the two spheres that was evident in period 1. Further, IJPDLM, TJ and JBL all appear to be located in the center of the network. This

JAT

IJPDLM

JBL T

IMM DS IJPMM JM

TJ

JMR

JTEP

TQ

LTR

MS NRL

Figure 2. Period 1 MDS mapping

IJPDLM 28,5 JBL

DS

IJPDLM

IMM

340

TJ NRL

LTR IJPMM JM

MS

TQ JTEP JAT

JMR T

Figure 3. Period 2 MDS mapping

indicates that these journals communicate with a very interdisciplinary set of journals. The second multivariate procedure employed was Johnson’s hierarchical clustering technique (Borgatti et al., 1992). This procedure is designed to find a series of nested partitions of the journals. The two most similar items are clustered together and, in an iterative procedure, other similar journals are consecutively added to the existing cluster such that, in the end, one single cluster of all the journals in the network emerges (Borgatti et al., 1992). In decreasing order of similarity, the analysis provides a hierarchical account of clustering. This procedure found that in period 1, the following pairs of journals were most similar to one another: JMR-JM, MS-NRL, JBL-LTR, and JTEP-T (see Figure 4). Period 2 results were fairly consistent with the results of period 1. For period 2, the following pairs were most similar: JMR-JM, MS-NRL, JBLIJPDLM, LTR-TJ and JTEP-T (see Figure 5). The two multivariate procedures used here provided a visual depiction of the relative position of the journals within the network. The results of these analyses confirm earlier results that find a much more cohesive network in period 2 as compared to period 1. These results also demonstrate that two primary journals in business logistics and transportation, JBL and TJ, were relatively close to the center of the network in period 1. IJPDLM joined them and together, the journals which represent the core of the fields of business logistics and transportation occupied an important central location in period 2. In terms of similarity clustering, it is not surprising to find that marketing journals were most similar to one another in both time periods, as were logistics and transportation journals.

I J P D I J L M M J M M R M

I J P N J L J M D M R B T T A M S S L L R J T

J T E T P T Q

A social network analysis

Level 4.3479 3.7738 2.7515 1.4590 1.2658 1.1583 0.8254 0.8021 0.6236 0.6017 0.5242 0.2320 0.2314 0.0528

. . XXX . . . . . . . . . . . . . XXX . . XXX . . . . . . . . XXXXX . . XXX . . . . . . . . XXXXX . . XXX . . . . XXX . . XXXXX . . XXX XXX . . XXX . . XXXXX . XXXXX XXX . . XXX . . XXXXX . XXXXX XXX . . XXXXX . XXXXXXX XXXXX XXX . . XXXXX . XXXXXXX XXXXX XXXXX . XXXXX X X XXXXXXX XXXXX XXXXX . XXXXX X X XXXXXXX XXXXX XXXXX XXXXXXX X X XXXXXXX XXXXX XXXXXXXXXXXXX X X XXXXXXXXXXXXX XXXXXXXXXXXXX X X XXXXXXXXXXXXXXXXXXXXXXXXXXX

Note: It is most useful to start at the top of the figure to identify the most similar journals and then move down the figure to see what other journals are next most similar. For example, JMR and JM are the most similar in this period. Then, IMM is the next most similar journal to those two. Level refers to the degree of similarity among journals in the clusters. At level 3.7738, for example, a journal is on average 3.7738 units similar to the other journal in the cluster.

Position in the network The first measure we used to assess position in the network was profile similarity. This measure was used in order to determine which journals were most similar to one another in terms of their communications patterns. Using Johnson’s profile similarity procedure, UCINET correlates rows and columns of each pair of journals and then clusters them according to degree of similarity (Borgatti et al., 1992). For two journals to be defined as similar, they must be more similar to one another than they are to any other journals in the network. The advantage of this procedure over the multivariate procedures is that this procedure does not symmetrize the data matrix, it allows the matrix to maintain all the information on the two-way communication flows. UCINET first identifies two very similar journals and then, in an iterative procedure, adds the next most similar journals to the first two, etc. For period 1, TJ and TQ are the two journals in the network that are most similar to one another. Next, T is identified as most similar to TJ and TQ. Finally, JAT and LTR are very similar to one another in the network. Taken together with the results of previous analyses, this analysis confirms that the transportation-related journals in this period are very similar to one another. This result is not entirely surprising because these journals accept articles that

341

Figure 4. Period 1 hierarchical clustering

IJPDLM 28,5

I J N I J P D M R M J M M S S L M M R M

I J P J D J L J T L B T T A E T M L R J T P T Q

Level

342

Figure 5. Period 2 hierachicial clustering

6.0363 4.0253 2.5347 2.3713 2.3255 1.5724 1.4313 1.3722 0.5421 0.4209 0.3737 0.3419 0.2384 0.1650

. . . . XXX . . . . . . . . . . XXX . XXX . . . . . . . . . . XXX . XXX . XXX . . . . . . . XXX . XXX . XXX XXX . . . . . XXX XXXXX . XXX XXX . . . . X X XXX XXXXX . XXX XXX . . . . X X XXX XXXXX . XXX XXX . XXX . X X XXX XXXXX . XXXXXXX . XXX . X X XXX XXXXXXX XXXXXXX . XXX . X X XXX XXXXXXXXXXXXXXX . XXX . X X XXX XXXXXXXXXXXXXXX XXXXX . X X XXX XXXXXXXXXXXXXXX XXXXXXX X X XXXXXXXXXXXXXXXXXXX XXXXXXX X X XXXXXXXXXXXXXXXXXXXXXXXXXXX

Note: It is most useful to start at the top of the figure to identify the most similar journals and then move down the figure to see what other journals are next most similar. For example, JMR and JM are the most similar in this period. Then, IMM is the next most similar journal to those two. Level refers to the degree of similarity among journals in the clusters. At level 6.0363, for example, a journal is on average 6.0363 units similar to the other journal in the cluster.

primarily deal with transportation-related issues and thus tend to exchange information among themselves, rather than exchange information with the broader network of journals who less often specifically discuss transportation topics. Period 2 again finds that transportation and logistics-related journals are similar to one another. The groupings of similar journals are as follows: JAT-TQ-T, DS-NRL, and LTR-TJ. This analysis reveals a somewhat broader grouping of different types of journals that are similar to one another within the network. The second measure of position is that of structural equivalence. Structural equivalence is a measure of the extent to which one or more journals share the same role or perform the same function within a given network. Structurally equivalent journals occupy the same social network positions in that they have connections with the same “other” journals (Scott, 1991). The output of this procedure represents commonalities in the roles that these journals play within the network. Although two journals may not cite one another very frequently, they may still perform the same type of role within the network by, for example, channeling information between different clusters of journals or collecting information from subsets of journals. The roles performed by these members can be many and varied. Further, structural equivalence is thought to be a

strong determinant of contagion (Burt, 1987), or the extent to which a new idea, A social network methodology, etc. is spread throughout a scientific community. analysis One way UCINET analyzes structural equivalence is through a procedure called CONCOR (“CONvergence of iterated CORrelations” (Scott, 1991)) which performs a factor analysis on the correlations of the distance matrix of the data (Borgatti et al., 1992). This analysis further supported the placement, or 343 position, of journals in the network by identifying journals that play similar roles within the network. Journals that are found to be structurally equivalent may be interchangeable with one another. In other words, they perform similar roles within the network. Period 1 results identify four blocks (of between one and eight journals) of structurally equivalent journals (see Table V). The first block contains DS, IMM, IJPDLM, and NRL. The second block contains only one journal: JBL. The third block contains JAT and LTR, while the fourth block contains a mix of transportation, management, and marketing journals: IJPMM, JM, JMR, MS, JTEP, T, TJ, and TQ. The period 2 analysis also produced four separate blocks of structurally equivalent journals, however, the journals within the blocks were very different from those in the first period. The first block contains DS, NRL, and JAT. The second block included IJPDLM, JBL, and LTR. The third block contained IMM, IJPMM, T, TJ, JTEP, and TQ. The final block included MS, JM, and JMR. The journals in the first block could be considered to be a motley mix of management science, logistics and transportation, the journals in the second block could be defined as logistics management journals, the journals in the third block could be defined as logistics and transportation management, and the journals in the fourth block could be defined as marketing and management science. With the exception of the mix of journals in the first block, this period seems to be characterized by blocks of journals that share a broad topic area. Thus, in contrast to period 1, period 2 contains a more coherent set of journals in each block. Theoretically, journals in any given block perform the same function, or role, within the network and should be interchangeable with one another. Scott (1991) admits that it is very difficult to define the exact roles that may be played by members in each block in this analysis. However, period 2 has four blocks and four separate and distinct roles that are played by member journals.

Block 1 Block 2 Block 3 Block 4

Period 1

Period 2

DS, IMM, IJPDLM, NRL JBL JAT, LTR IJPMM, JM, JMR, MS, JTEP, T, TJ, TQ

DS, NRL, JAT IJPDLM, JBL, LTR IMM, IJPMM, T, TJ, JTEP, TQ MS, JM, JMR

Table V. Structural equivalence blocking

IJPDLM 28,5

344

The two analyses of position in the network identify those journals that are structurally similar to one another and perform similar roles within the network. These analyses indicate that the roles of the journals within each block become more clearly defined in period 2, in part, because they are grouped by a broad topic area. Journals within a particular block work together to perform a particular role and it is easier for them to do this in period 2 because journals are grouped with other journals that share an interest in the same general subject matter. Discussion Using four broad social network analyses on the citation network – measures of centrality, a measure of groupings, multivariate measures, and position in the network – the primary goals of this research were successfully completed. First, a wide variety of social network analysis techniques were discussed and applied to citation analysis in an examination of a scientific community. Second, those social network analysis techniques were applied to the network of business logistics and transportation journals to better understand how, in this case, a scientific discipline evolves and develops into its own distinct, wellestablished scientific community. Numerous studies suggest that location in a network (Zinkhan et al., 1992), a “research front” (Price, 1965), or a central position in the network (Burt, 1982; 1987; Coleman et al., 1957; Crane, 1972; Knoke and Burt, 1983) is critically important in the adoption and dissemination of ideas. In general, the members of the present network that may perform this role are JM, JMR, MS, IJPDLM, JBL, and TJ. These journals seem to consistently be identified by the different social network analyses as journals that play an important role in disseminating knowledge throughout the network. These journals can be considered to be the “invisible college” (Crane, 1972), or the gatekeepers of knowledge in the business logistics and transportation scientific community. While the first period was characterized by several key journals playing important gatekeeper roles, period 2 was characterized by more direct communication between journals. Thus, rather than a stratified hierarchy of important journals dictating the flow of information, communication in this network now is much more open and direct. Rather than journals preempting one another in an attempt to become more powerful or prestigious in this network, the business logistics and transportation journals appear to be sharing power with, rather than usurping power from, previously prestigious journals. The roles each of the journals performed in helping make the network operate more efficiently is an important consideration in concluding whether or not the field emerged in order to fulfill a distinct need in the network (Berry and Parasuraman, 1993). Taken together, the foregoing analyses indicate that the appearance of the network changed considerably between period 1 and period 2. In general, period 1 is characterized by a very distinct break between logistics and transportation (see Figure 2). The journals in this period do not directly

communicate with one another and instead rely on other journals to perform A social network the role of gatekeeper by funneling information to other parts of the network. analysis Groupings of journals are haphazard and contain a confusing mix of very different journals. The appearance of the network in period 2 is much more organized. The flow of information between journals is more efficient in that journals directly communicate with one another rather than relying on other 345 journals to perform the gatekeeper role. Further, there is no longer a distinct break between logistics and transportation. Instead, the business logistics and transportation journals are located very close to the center of the network (see Figure 3). In addition, groupings of the journals are much more sensible in that they are grouped, in general, by broad topic area. We can thus conclude that the network was much more efficient and orderly in period 2. We find that Business Logistics and Transportation evolved in such a way that it fulfilled an important communication need within the network. In general, period 1 is characterized by a need for more easy and efficient communication between member journals. This need is fulfilled in period 2 where JBL, TJ, and IJPDLM locate themselves close to the center of the network and ensure that information flows are much more direct and open between members. Interestingly, IJPDLM plays an especially important role in the network because it is the journal that is the most “between” other journals. Because the network takes on this structure in period 2, the flow of important new innovations and ideas through the network is much more rapid and efficient. The evolution of business logistics and transportation throughout time makes for an interesting case study. The ten years separating period 1 and period 2 encompassed a wide range of macro-environmental factors that likely affected the structure of the communication patterns in this scientific field. What may happen in the next ten years is likely to have an even more radical impact on the way scholars communicate. The Internet allows for the instantaneous transfer of ideas to scholars all over the world. Therefore, rather than wait for the arrival of their quarterly publications, researchers can access on-line versions of journals, download copies of articles, and collaborate with other scholars all over the world. Will citation analysis become obsolete in the next decade? It is impossible to speculate. What we can say, however, is that the advancements that will take place will provide an even more fascinating forum within which we can continue to study the ongoing evolution and maturation of scientific disciplines. Notes 1. In selecting the journals, we attempted to develop as comprehensive a list of academic journals that were related to business logistics and transportation as possible. We acknowledge that other researchers may take issue about why a particular journal was included or not included in our network. However, we believe the more interesting question is how this set of journals behaves over time. 2. All analyses were performed using the social network analysis program UCINET IV (Borgatti et al., 1992). This computer program uses matrices as input and performs a

IJPDLM 28,5

346

comprehensive selection of most social network analyses. With its flexible menu-driven interface, the user has numerous options to specify specific matrix manipulations and types of analyses. 3. An examination of the period 1 matrix reveals that this result is most likely due to a strong one-way flow of information from DS and MS to JBL. JBL makes frequent citations to DS and MS, but DS and MS do not cite JBL in this time period. References Allen, B.J. and Vellenga, D.B. (1987), “Affiliation of authors in transportation and logistics academic journals – an update”, Transportation Journal, Vol. 26 No. 3, pp. 39-47. Bartels, R. (1965), “Development of marketing thought: a brief history”, in Schwartz, G. (Ed.), Science in Marketing, John Wiley, New York, NY. Berry, L.L. and Parasuraman, A. (1993), “Building a new academic field – the case of services marketing”, Journal of Retailing, Vol. 69 No. 1, Spring, pp. 13-60. Borgatti, S., Everett, M.G. and Freeman, L.C. (1992), UCINET IV Version 1.0, Analytic Technologies and Reference Manual, Columbia. Borgman, C.L. (1990), Scholarly Communication and Bibliometrics, Borgman, C.L. (Ed.), Sage, Newbury Park, CA. Bowersox, D.J. (1966), “Physical distribution in semi-maturity”, Air Transportation, January, pp. 9-11. Burt, R.S. (1982), “Stratification in elite sociological methodology”, in Toward a Structural Theory of Action, Academic Press, New York, NY. Burt, R.S. (1987), “Social contagion and innovation: cohesion versus structural equivalence”, American Journal of Sociology, Vol. 92, pp. 1287-335. Coleman, J., Katz, E. and Menzel, H. (1957), “The diffusion of innovation among physicians”, Sociometry, Vol. 20, pp. 253-70. Cote, J.A., Leong, S.M. and Cote, J. (1991), “Assessing the influence of journal of consumer research: a citation analysis”, Journal of Consumer Research, Vol. 18, December, pp. 402-10. Crane, D. (1972), Invisible Colleges: Diffusion of Knowledge in Scientific Communities, The University of Chicago Press, Chicago, IL. Doreian, P. (1985), “Structural equivalence in a psychology journal network”, Journal of the American Society for Information Science, Vol. 36 No. 6, pp. 411-17. Doreian, P. (1988), “Testing structural-equivalence hypotheses in a network of geographical journals”, Journal of the American Society for Information Science, Vol. 39 No. 2, pp. 79-85. Emmelhainz, L.W. and Stock, J.R. (1989), “An evaluation of logistics and related journals”, International Journal of Physical Distribution and Materials Management, Vol. 19 No. 12, pp. 40-5. Everett, J.E. and Pecotich, A. (1993), “Citation analysis mapping of journals in applied and clinical psychology”, Journal of Applied Social Psychology, Vol. 23 No. 9, pp. 750-66. Fawcett, S.E., Vellenga, D.B. and Truitt, L.J. (1995), “An evaluation of logistics and transportation professional organizations, programs, and publications”, Journal of Business Logistics, Vol. 16 No. 1, pp. 299-314. Ferguson, W. (1983), “An evaluation of journals that publish business logistics articles”, Transportation Journal, Vol. 22 No. 4, Summer, pp. 69-72. Gatrell, A.C. (1984), “Describing the structure of a research literature: spatial diffusion modeling in geography”, Environment and Planning B: Planning and Design, Vol. 11, pp. 29-45. Hamelman, P.W. and Mazze, E.M. (1973), “Cross-referencing between AMA journals and other publications”, Journal of Marketing Research, Vol. 10, May, pp. 215-8. Hoffman, D.L. and Holbrook, M. (1993), “The intellectual structure of consumer research: a bibliometric study of author cocitations in the first 15 years of the journal of consumer research”, Journal of Consumer Research, Vol. 19, March, pp. 505-17.

Iacobucci, D. and Hopkins, N. (1992), “Modeling dyadic interactions and networks in marketing”, Journal of Marketing Research, Vol. 24, February, pp. 5-17. Knoke, D. and Burt, R.S. (1983), “Prominence”, in Burt, R.S. and Miner, M.J. (Eds), Applied Network Analysis: A Methodological Introduction, Sage, Beverly Hills, CA. Leong, S.M. (1989), “A citation analysis of the journal of consumer research”, Journal of Consumer Research, Vol. 15, May, pp. 492-7. Noma, E. (1983), “Untangling citation networks”, Information Processing & Management, Vol. 18 No. 2, pp. 43-53. Phillips, D.M., Baumgartner, H. and Pieters R. (1998), “Position and influence in the evolving citation network of the Journal of Consumer Research”, working paper. Price, D.J. (1965), “Networks of scientific papers”, Science, Vol. 149, 30 July, pp. 510-14. Reed, K.L. (1995), “Citation analysis of faculty publication: beyond science citation index and social science citation index”, Bulletin of the Medical Library Association, Vol. 83 No. 4, October, pp. 503-8. Reingen, P.H. and Kernan, J.B. (1986), “Analysis of referral networks in marketing: methods and illustration”, Journal of Marketing Research, Vol. 23, November, pp. 370-8. Rice, R.E., Borgman, C.L., Bednarski, D. and Hart, P.J. (1989), “Journal-to-journal citation data: issues of validity and reliability”, Scientometrics, Vol. 15 No. 3-4, pp. 257-82. Salancik, G.R. (1986), “An index of subgroup influence in dependency networks”, Administrative Science Quarterly, Vol. 31, June, pp. 194-211. Scott, J. (1991), Social Network Analysis: A Handbook, Sage, London. Shaw, A.W. (1916), An Approach to Business Problems, Harvard University Press, Cambridge, pp. 703-65. Zinkhan, G.M., Roth, M.S. and Saxton, M.J. (1992), “Knowledge development and scientific status in consumer-behavior research: a social exchange perspective”, Journal of Consumer Research, Vol. 19, September, pp. 282-91.

A social network analysis

347

Appendix IJPIJPDS IMM DLM JAT JBL JM JMR MM JTEP LTR MS NRL DS IMM IJPDLM JAT JBL JM JMR IJPMM JTEP LTR MS NRL T TJ TQ

4.39 0.12 0.31 0.00 0.90 0.17 0.40 0.34 0.00 0.09 0.94 0.06 0.00 0.06 0.00

0.00 5.10 0.46 0.00 0.36 0.27 0.17 1.70 0.00 0.00 0.02 0.00 0.00 0.00 0.00

0.00 0.29 2.67 0.00 1.81 0.15 0.02 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00

0.00 0.00 0.00 1.26 0.00 0.00 0.00 0.00 0.00 0.19 0.00 0.00 0.11 0.06 0.00

0.00 0.18 0.26 0.00 0.90 0.06 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.25 0.00

0.31 1.75 0.06 4.63 5.63 1.23 1.34 1.59 1.08 0.00 0.00 0.00 2.35 0.18 0.36 7.81 5.99 0.06 2.70 14.25 0.02 2.38 0.51 13.41 0.00 0.00 0.00 0.00 0.00 0.00 0.29 1.32 0.00 0.00 0.06 0.00 0.00 0.33 0.00 0.06 0.00 0.06 0.00 0.00 0.00

0.00 0.00 0.15 2.24 0.00 0.04 0.00 0.00 7.55 1.86 0.00 0.00 1.76 0.62 1.00

0.00 0.12 0.26 0.28 2.53 0.00 0.00 0.00 0.10 1.67 0.01 0.00 0.22 0.37 0.18

5.21 1.99 1.44 0.28 5.42 1.35 1.05 0.34 0.00 0.46 9.25 6.75 0.00 0.06 0.00

0.34 0.00 0.00 0.00 0.54 0.00 0.00 0.00 0.00 0.00 0.80 7.18 0.00 0.06 0.00

T

TJ

TQ

0.00 0.00 0.00 1.12 0.00 0.02 0.02 0.00 1.16 0.19 0.00 0.00 3.62 0.06 0.18

0.00 0.12 0.10 0.00 1.08 0.00 0.00 0.51 0.10 0.84 0.00 0.00 0.11 2.64 0.27

0.00 0.00 0.00 0.70 0.00 0.04 0.00 0.00 0.29 0.46 0.00 0.00 1.65 0.43 2.83

Table AI. Period 1: normalized matrix

IJPDLM 28,5

348

Table AII. Period 2: normalized matrix

IJPIJPDS IMM DLM JAT JBL JM JMR MM JTEP LTR MS NRL DS IMM IJPDLM JAT JBL JM JMR IJPMM JTEP LTR MS NRL T TJ TQ

4.89 0.13 0.76 0.00 1.11 0.06 0.08 0.00 0.08 0.56 0.41 0.83 0.00 0.18 0.00

0.05 7.12 1.27 0.00 0.62 0.47 0.32 0.72 0.00 0.06 0.05 0.00 0.00 0.48 0.00

0.01 0.20 3.06 0.00 2.63 0.03 0.02 0.36 0.00 1.05 0.00 0.00 0.00 1.25 0.06

0.00 0.00 0.00 1.44 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.07 0.00 0.56

0.06 0.00 2.44 0.00 7.05 0.02 0.00 0.27 0.00 1.73 0.00 0.10 0.00 1.55 0.06

0.79 1.03 0.20 6.11 3.36 0.40 1.79 0.96 3.13 0.00 0.00 0.00 1.31 0.97 1.31 8.79 6.55 0.10 5.52 11.84 0.08 1.34 0.81 14.96 0.08 0.15 0.00 0.37 0.25 0.37 0.25 0.58 0.03 0.05 0.05 0.23 0.22 0.44 0.00 0.48 0.30 0.42 0.34 0.11 0.06

0.01 0.00 0.14 1.57 0.00 0.00 0.02 0.00 5.98 1.60 0.00 0.00 2.03 0.36 0.34

0.02 0.00 0.45 0.00 0.97 0.00 0.00 0.09 1.29 3.52 0.00 0.00 0.44 1.97 0.23

6.59 0.84 1.75 0.92 2.35 1.35 2.36 0.63 0.15 0.86 8.39 7.22 0.07 0.42 0.00

0.39 0.00 0.07 0.13 0.07 0.00 0.00 0.09 0.00 0.12 0.83 6.14 0.00 0.00 0.00

T

TJ

TQ

0.00 0.00 0.03 0.26 0.00 0.00 0.00 0.00 0.83 0.19 0.00 0.00 3.41 0.06 0.34

0.00 0.07 1.10 0.26 1.73 0.02 0.00 0.00 0.53 2.78 0.00 0.00 0.15 5.84 0.56

0.00 0.00 0.10 0.13 0.14 0.00 0.00 0.00 0.30 0.19 0.00 0.00 0.36 0.66 2.14

Note The matrices are normalized such that the sum of all cell entries is equal to zero. The cell entries, therefore, can be likened to percentages. Rows indicate sending patterns while columns indicate receiving patterns. In period 1, for example, the citations made by IMM to DS account for 0.12 per cent of the total citations in the matrix. Conversely, the citations made by DS to IMM account for 0 per cent of the total citations in the matrix.