Significant Clustering: Implications for Strategic Groups Research

Firms following similar strategies bump into each other while attracting suppliers and buyers. This triggers an awareness of their mutual interdependence and creates incentives for competition among similar firms. If groups are sufficiently isolated, cut-throat competition within one group would not spread to other groups. Thus, pockets of oligopolistic competition could emerge within the industry and create the potential for performance differences. A significance test is needed to determine if firms clump together to form isolated strategic groups (Hatten and Hatten, 1987; Barney and Hoskisson, 1990). Two complementary significance tests are illustrated: a Monte Carlo test and a permutation test (Clarke, Somerfield and Gorley, 2008). A hierarchical cluster analysis is performed using Ward‟s method and squared Euclidean distance to identify strategic groups in the EU airlines industry (n=26).A permutation test and a Monte Carlo test are used to assess the number of significant clusters. Performance differences across groups on ROA were assessed using a one-way ANOVA. The solutions for 2-12 clusters were significant and the solutions for 2, 3, and 6 (hierarchically nested) clusters were quantitatively attractive. The 6 cluster solution offered the most intuitively appealing groupings. A one-way ANOVA found significant differences in ROA across the six groups. In the absence of a significance test for clustering, differences in performance became the litmus test for the existence of the groups themselves. Consequently, groups became defined by the barriers that protect them (Harrigan, 1985; McGee & Thomas, 1986). This conflated the definitions for these closely related but logically distinct concepts. The advent of significance tests removes this constraint. Theorists are now free to model the concepts of strategic groups and mobility barriers separately and to examine the intriguing interplay between them. Monte Carlo studies are currently underway to assess the Type I and Type II error rates of these significance tests under a wide range of conditions.