9Why Do MNCs Divest or Retain Foreign Subsidiaries?Approaches from Dependency and Redundancy in Subsidiary Networks Naoki YasudaControl variables I included control variables for each subsidiary, its parent firm, and host country. I refer to Berry (2013) and Dai, Eden, and Beamish (2013) to select control variables. I incorporated equity from Japan to control for the influence of a parent firm. The previously described local sales ratio and local procurement ratio are used to control for subsidiaries’ activities in host countries. To control for subsidiary size, I incorporated the number of employees in the subsidiary. Return on sales, computed as the ordinary profit of each subsidiary divided by total sales, indicates the subsidiary’s profitability. I control for a subsidiary’s research and development intensity by dividing the subsidiaries’ R&D expenses by total sales. To control for type of subsidiary, I incorporate three dummy variables for the following: export platform, vertical type, and networked type subsidiaries. The parent-level variables include international experience, size, research and development activities, and profitability. Parent profitability is defined by parent firm ROS (i.e., ordinary profits of subsidiaries divided by total sales). The size of an MNC is defined by its number of employees. I calculated the research and development intensity of an MNC by dividing the MNC’s R&D expense by total sales. To control for MNC international experience, I identified when the MNCs began operating abroad and measure how long these firms remain in business. Host country characteristics including country growth, country maturity, and policy stability can influence MNCs’ divestment decisions (Berry, 2013). I derived a country’s annual percentage GDP growth from annual World Development Indicators from the World Bank. I used GDP per capita to control maturity of the host countries (e.g., Xia, 2011). I used data from POLCON to measure political risks, a political constraint variable developed by Henisz (2000) and widely used in international relations research (e.g., Holburn & Zelner, 2010). Statistical Analysis To predict the likelihood of divestment, I followed previous studies and used a Cox proportional hazards model (e.g., Kang, Lee, & Ghauri, 2017). “This is one of the most widely used methods of survival analysis and estimates the effects of different covariates that influence time to failure” (Berry, 2013: 254). I used STATA14 for the estimation using the “xtcox” command with the “vce (robust)” option. I lessened the effects of unobserved heterogeneity by incorporating year-dummy variables to control for yearly fixed effects.RESULTS Table 2 presents descriptive statistics and correlations. Although some coefficients are relatively high, the highest variance inflation indicator (VIF) was below 10 and the average is 2.76. Therefore, I took no remedial action to address multicollinearity. Using divestment of a subsidiary as the dependent variable, Table 3 shows the results of event history analysis. Model 1 tests the effects of control variables. Divestment is less likely when equity
元のページ ../index.html#13