臺大管理論叢 NTU Management Review VOL.30 NO.3

The Relationships among Characteristics of Interlocking Directorate Network, Technological Diversity and Innovation Performance: Evidence from Taiwan’s Electronics Industry 172 the mediation effect of technological diversity on the relationships between these networks and innovation performance. 2. Design/Methodology/Approach Taiwan’s electronics industry is a major global player with a wide range of knowledge-intensive characteristics. This study uses 406 listed electronics firms in Taiwan as the research sample with 7 years observation (from 2010 to 2016). After deleting incomplete data, the final sample size is 2,807 observations. All the 406 Taiwanese electronics firms file patents with the U.S. Patent and Trademark Office (USPTO) during our observation period. We choose the filing as a measure of innovation performance for two reasons. First, the processes of applying U.S. patents are more complicated and much stricter than the applying processes in Taiwan or other countries. Second, the U.S. is one of the major exporting markets of Taiwan’s electronics industry. These Taiwanese firms value the importance of American market, which reflect on applying the U.S. patents. (Chin, Chen, Kleinman, and Lee, 2009). Therefore, we believe our measurement of innovation performance of these firms is representable. Following Kim, Lee, and Cho (2016) approach of measuring technological diversity, we classify the applied U.S. patents from aforementioned firms using the International Patent Classification (IPC), which divides these patents into 23 technical fields and 634 sub-categories. Furthermore, we collect data about directors’ network characteristics from the Shareholdings of Directors of Corporate Governance Board Database hosted by the Taiwan Economic Journal (TEJ). We then utilize UCINET software to calculate the directors’ network characteristics. Finally, we obtain the financial information of our targeted firms from the Finance Database of TEJ. As for the mediation effect, early studies commonly use the causal steps approach outlined in Baron and Kenny (1986) to investigate the mediation. Nevertheless, Hayes (2009) proposes that while this method derives the causal relationship and draws conclusions logically, it does not directly analyze the mediating effect, making it difficult to avoid type 1errors. Therefore, in this study, we follow Preacher and Hayes (2008) and Hayes (2009) instead. We then employ Stata for Structural Equation Modeling (SEM) and bootstrapping with 1,000 reiterations to produce estimates of indirect effects and their