7Why Do MNCs Divest or Retain Foreign Subsidiaries?Approaches from Dependency and Redundancy in Subsidiary Networks Naoki Yasudasubsidiaries with same functions at a lower cost. Reducing redundancy contributes to the efficiency of subsidiary networks. As a result, MNCs will more likely divest subsidiaries that duplicate functions because doing so reduces network redundancy. Hypothesis 3a: The redundancy of subsidiary functions in the same country where a focal subsidiary exists has a positive effect on an MNC’s subsidiary divestment. Hypothesis 3b: The redundancy of subsidiary functions in the same region where a focal subsidiary exists has a positive effect on an MNC’s subsidiary divestment. METHODResearch Setting Testing hypotheses requires subsidiary-level data for local sales, procurement, imports, and exports. This study obtained subsidiary-level data about Japanese foreign affiliates from the Japanese Ministry of Economy, Trade, and Industry (METI) survey, an annual survey of all Japanese affiliates worldwide. The quality of this data has been widely endorsed by existing studies (e.g., Makino, Chan, Isobe, & Beamish, 2007). Data and Sample The METI survey covers the period 1995–2013 and reports the names of subsidiaries divested in each year. To reveal longitudinal dynamics, I used data from the entire time frame. I excluded firms with no foreign subsidiaries, those that lacked the necessary data, those with return on sales exceeding 1, and parent firms with return on sales exceeding 1. I also excluded any subsidiaries whose parent company is not in manufacturing. The final sample covered 8,139 subsidiaries in 83 countries and comprised 44,228 data points (unit of analysis is subsidiary-year).MeasurementDependent variables Divestment of subsidiaries is the dependent variable. To construct the dependent variables and test the hypotheses, I follow Berry (2013). A binary variable is coded 1 if an MNC divests a foreign affiliate, or 0 if no divestment occurs that year. I used time t + 3 for all dependent variables because divestment is often a time-consuming activity. I set the dependent variable time window from t + 1 to t + 3 for a robustness check.Independent variables I used time t data for all independent and control variables. Independent variables capture the network effects of divesting a subsidiary. To test Hypotheses 1a and 2a, I constructed hub functions for country and regional networks.