Subsidiary Types, Activities, and Location: An Empirical Investigation

A restricted maximum likelihood estimation procedure (REML) was employed to estimate themultilevel random effects mixed logit model.Categorical modeling results: Table 3 reports the results for the multilevel randomeffects mixed model logit analysis for subsidiary type with the firm random effect and the hosteconomy fixed effect.TABLE 3 GOES ABOUT HEREAll of the estimates are generated simultaneously, but are placed into columns by type for clarity.Thus the columns all form part of the same model estimate, rather than separate estimates bytype. Each estimate reflects that probability that a subsidiary will be of a certain type relative tothe probability it will be the type represented by the baseline response, which is the “Sales andService” subsidiary.FIRMSIZE has a strong, positive influence on the probability that a subsidiary will be a“Management and Development Subsidiaries”, “Full Functional Subsidiaries”, “ProductionBases”, and “Sales and Service Subsidiaries”. Particularly strong results are found for the “FullFunctional Subsidiaries” category. Thus hypothesis H1 is supported. The international experienceof the firm has in the Asia-Pacific has a significant influence only on the subsidiary type“Production Bases”. This may indicate that investments in production are influenced by priorexperience of managing geographically dispersed production. Hypothesis H2 is partiallysupported.The subsidiaries of Japanese firms are more likely to be involved in activities that consist of“Management and Development” and “Production Bases” and less likely to be “Full Functional23