Mapping the Dynamics of Strategic Alliance Networks in the Global Information Sector: David Knoke University of Minnesota Workshop on Clusters, Networks & Alliances in the Telecommunication Sector School of Management, University of Surrey June 11, 2003 Mapping the Dynamics of Strategic Alliance Networks in the Global Information Sector


Lost in Space? : Lost in Space? When constructing maps of social spaces for interorganizational relations, analysts face crucial decisions about optimal procedures for: 1: Measuring distances or proximities between pairs of organizations based on interactions (and avoidances?) 2: Locating organizations within multidimensional spaces that accurately represent these distances/proximities 3: Identifying which organizations jointly occupy positions (subgroups) based on equivalence/similarity of their interorganizational relations Proposed solutions combine methods drawn from numerical taxonomy, classification, graph theory, multidimensional scaling, and cluster analysis.


The Global Info Sector: The Global Info Sector To illustrate these choices, I will analyze data on the announced strategic alliances among the world’s largest info firms, the Global Information Sector: Five NAIC information subsectors (publishing; motion pictures & sound recording; broadcasting & telecomms; info services & data processing) plus the computer, telecommunications, and semiconductor manufacturing industries 145 organizations: 66% USA, 16% Europe, 15% Asia Alliances & ventures announced in general and business news media from 1989 to 2000 Total of 3,569 strategic alliances involving two or more orgs (many collaborations also include additional partners)


Strategic Alliances: Strategic Alliances Strategic alliance: at least two partner firms that (1) remain legally independent; (2) share benefits, managerial control over performance of assigned tasks; (3) make contributions in strategic areas, e.g., technology or products (Yoshino & Rangan 1995) SA governance forms vary in the types of legal and social mechanisms to coordinate & safeguard alliance partners’ resource contributions, administrative responsibilities, divide rewards from their collaboration (Todeva & Knoke 2003) Hierarchical Relations --------------------------------------------------------- JOINT VENTURES COOPERATIVES EQUITY INVESTMENTS R&D CONSORTIA STRATEGIC COOP. AGREEMENTS CARTELS FRANCHISING LICENSING SUBCONTRACTOR NETWORKS INDUSTRY STANDARDS GROUPS ACTION SETS --------------------------------------------------------- Market Relations Todeva, Emanuela and David Knoke. 2003. “Strategic Alliances and Corporate Social Capital.” Kölner Zeitschrift für Sociologie und Socialpsychologie (Forthcoming) Yoshino, Michael Y. and U. Srinivasa Rangan. 1995. Strategic Alliances: An Entrepreneurial Approach to Globalization. Cambridge, MA: Harvard University Press.


30 Core GIS Orgs: 30 Core GIS Orgs My examples use the alliances among the 30 most-active GIS firms in 2000, from three continents and a dozen industries


1: Measuring Distances: 1: Measuring Distances The social distance between a pair of actors varies with their interaction frequency and/or strength. A strategic alliance is an interaction event where orgs are present or absent as partners. Across all alliances, a 2x2 table displays the pattern of partnerships among each pair. In 2000, the GIS orgs announced 452 alliances; 209 involved at least two of the 30 core orgs. Here are the Microsoft-IBM counts, with the four cell frequencies denoted by conventional letters: a= d= b= c=


What Really Counts?: What Really Counts? For interval data, Euclidean distance and correlation are apt metrics. But for binary counts, several other measures are more appropriate. Some include both the joint presences and joint absences: Others exclude the joint absences: A basic question is whether to include or exclude the number of alliances involving neither organization? (Akin to the biological taxonomy decision to count or ignore “absence of feathers” when classifying fish species.) Co-absence probably doesn’t indicate mutual avoidance, because every org is not a plausible partner for most alliances. As all orgs participate in a minority of events, cell “d” carries heavy weight, which argues strongly for using distance/similarity measures that exclude a pair’s joint absences.


2: Plotting Locating: 2: Plotting Locating Using the distances/proximities among all organizations, multidimensional scaling (MDS) programs can plot their relative locations in 2-, 3-, or higher-dimensional spaces. Input is a square, symmetric matrix whose entries measure the similarity/dissimilarity or equivalence between each row-and-column pair; main diagonal entries are set to 0. An MDS program represents a pair of organizations that is more proximate in the input data as located closer in the multidimensional space; less-proximate pairs are located farther apart. MDS outputs estimated spatial coordinates for N-dimensions. Using these coordinates, org locations are displayed in a diagram. A stress value summarizes how well the estimated locations fit the observed input data; lower stress (< .20) indicates a better fit. Kruskal, J.B. and M. Wish. 1978. Multidimensional Scaling. Newbury Park, CA: Sage. Schiffman, S.S., M.L. Reynolds and F.W. Young. 1981. Introduction to Multidimensional Scaling: Theory, Methods, and Applications. New York: Academic Press.


3: Encircling Positions: 3: Encircling Positions Analysts can cluster analyze the distance/proximity matrix to identify the organizations that jointly occupy positions (subgroups) based on their similar/equivalent interorganizational relations. This information is used to draw contiguity lines that encircle the position members in a MDS diagram; with luck, the positions form tight circles, not sprawling amoebas with tangled pseudopods. Three general types of clustering algorithms, with multiple options for specifying data, distances, and linkage procedures: Hierarchical cluster analysis: divisive or agglomerative methods that find smaller organizational clusters nested inside larger ones; dendograms (tree diagrams) reveal these hierarchical connections. Nonhierarchical cluster analysis: (k-means) divisive methods for interval data that iteratively reallocate organizations among new sets of nonnested clusters until the N user-specified clusters emerge Fuzzy cluster analysis: instead assignments to sharply separated clusters, fuzzy methods specify a membership degree (from 0 and 1) showing how likely an organization belongs to each cluster


Comparing Maps: Comparing Maps The following two pairs of diagrams compare the MDS and hierarchical cluster results for two contrasting measures of the distances between the 30 core GIS organizations in 2000. The resulting maps differ in several key features: For SMC distances (which count cell “d”): Organizations creating the most alliances – Microsoft, IBM, and Japanese firms – are located at the periphery, but firms with few partners fill a dense central “black hole” For Jaccard distances (which ignore cell “d”): The most active American and Japanese organizations are nearer one another, and Microsoft (which formed the most partnerships) is smack in the center Jointly occupied positions: The U.S. computer organizations have a sprawling position in the SMC figure, but more tightly clustered in the Jaccard plot. The Asian firms are split across two positions in SMC, but altogether using Jaccard. Telecomms occupy distinct positions at the top of the latter map.


MDS with SMC: MDS with SMC (UCINET; stress = 0.19)


Hclus SMC: Hclus SMC (SPSS centroid clustering)


MDS with Jaccard: MDS with Jaccard (UCINET; stress = 0.17)


Hclus Jaccard: Hclus Jaccard (SPSS centroid clustering)


The Art of Organizational Cartography: The Art of Organizational Cartography Constructing maps of interorganizational and other types of social relations involves as much art as science. Researchers today enjoy an abundance of concepts, measures, and computer graphics programs. Unfortunately, no clearly superior mapping method fits every purpose. Decisions about which options to exercise rest on assumptions and simplifications that can distort as well as reveal underlying structures. Plotting and clustering algorithms are susceptible to local optima, which urges great caution and repetition of analyses before drawing conclusions. Most importantly, organizational cartographers should be guided by strong theoretical expectations about the relationships they investigate. Greater efforts must be undertaken to construct theories of organizational space with testable propositions. To do otherwise puts Descartes before des horse. Martin Waldseemüller’s Carta Mariana, which named the southern continent after Amerigo Vespucci.