Studies on the dynamic patterns of industry convergence using network analysis

Abstract

As technology has advanced and firms’ inter-industry activities have increases, the evidence of industry convergence becomes increasingly available. However, whether industry convergence is the ubiquitous phenomenon and what are the patterns of industry convergence are largely ignore in academics. For that reason, this dissertation aims to quantitatively study on the dynamic patterns of industry convergence using network analysis in the perspective of interorganizational dynamics. To achieve this objective, this dissertation mainly focuses on how to measure industry convergence and what to be cautious if it presents. The measurement of industry convergence is covered in the second chapter, where it is measured in terms of co-occurrence based text-mining method. To do so, text mining is conducted using a large volume of unstructured data―2 million newspaper articles from 1989 to 2012. With the data, this chapter presents industry convergence that is taking place in the entire U.S. industries, focusing on its trends and patterns. If industries are interdependent as they are converging together, then traditional methods for cross industry analysis may be biased as they mostly assume the independency of observations instead of interdependency among them. The third chapter concerns possible problems of using centrality indexes in ordinary least square analysis. With the network representation, many scholars calculate its centrality indexes and use them in the regression analysis. However, some regression models assume the assumption of uncorrelated error terms and independent observations. If network dependency are not controlled, the statistical significances of the centrality variables may arise as they rather try to fill the gap of the underlying network dependency. To investigate this confounding effect, the third chapter shows that the statistical significance of a centrality parameter decreases after interactions between different units are properly controlled. Spatial models are provided as one possible examining tool to consider these concerns.