Social structure and alliance formation patterns: a longitudinal analysis - includes appendix

Administrative Science Quarterly,  Dec, 1995  by Ranjay Gulati

• E-mail
• Print
• Link

I checked the groups' classifications with hierarchical clustering against the K-means clustering algorithm and further cross-checked them with multiple industry experts and against recent studies of similar industries (Nohria and Garcia-Pont, 1991). Each industry expert was given the list of industry participants on index cards and asked to group them into up to ten groups, with each group including firms with similar strategic and financial capabilities. There was over 90 percent convergence in the results obtained from the experts and those obtained with the clustering analysis. I discussed anomalies with the experts and made assignments based on these discussions. Through this partitioning analysis, I identified seven distinct groups in the new materials sector and nine groups each in the industrial automation and automotive sectors.

I used the groups created to construct the variable strategic interdependence, coded 1 if the firms in a dyad belonged to different niches and 0 if they belonged to the same niche.(2) Additional measures of strategic interdependence based on specific firm attributes (size, performance, solvency, and liquidity) were also included in the analysis and are discussed further with the control variables.

Constructing the Social Networks

To compute the social structural measures, I constructed adjacency matrixes representing the relationships between the actors in a network. Because my focus here was on alliances formed within industries, I computed separate matrixes for each industry. I computed matrixes for each industry for each year, including all alliance activity among industry panel members through the year prior to the given year. Additional data on alliances announced by the panel members between 1970 and 1980 were entered into the initial matrix for 1981. I entered all matrixes into UCINET IV, a versatile software package that allows the computation of various network measures (Borgatti, Everett, and Freeman, 1992).

In constructing the social networks of past alliances, I made three choices about how to treat alliances. The first relates to the treatment of different types of alliances. Alliances range from closely intertwined equity joint ventures, at one extreme, to arm's-length licensing agreements, on the other. Each type entails varying levels of organizational commitments and leads to differing levels of organizational interdependence. Thus it is difficult to justify treating all alliances identically. In constructing the adjacency matrices, I chose to weight each type of alliance on the basis of the strength of the resulting relationship. The weighting scheme, ranging from 1 (weak) to 7 (strong) was based on prior weighting schemes used in alliance research (Contractor and Lorange, 1988; Nohria and Garcia-Pont, 1991). To ensure the robustness of the findings, I tested the results against those obtained using a simple dichotomous matrix that treated all alliances as the same.

The second choice relates to the treatment of multiple ties between two firms over the observed time period. I identified three possible approaches: (1) using an additive measure yielding scores as firms make multiple ties, (2) adding the scores and normalizing them by the maximum score possible in that year, and (3) using a Guttman scale to capture the score of the strongest alliance the firms had formed. I used a Guttman scale for the final analysis but compared the results against those produced using the other two approaches.