Subsidiary Types, Activities, and Location: An Empirical Investigation

In the present analysis however, each firm “i” appears twelve times (once for each economy). Inaddition, each host economy “k” appears 440 times (once for each firm). Thus one cannotassume that all the ξ ijk terms are independent. In other words, one cannot assume that decisions offirm “i” whether or not to have a subsidiary of type “j” will be independent across all firms “i”or locations “k”. In such cases (cases of clustered data), ignoring the potential correlations withinfirms and within locations can produce standard errors that are underestimated, test statistics thatare overestimated, inefficient parameter estimates, and biased parameter estimates (Allison, 1991and Agresti, 2002). In order to deal with the potential correlations within each cluster, a randomeffects mixed logit model was employed.The model included explanatory variables and random effects for both firms and locations in aneffort to determine the explanatory factors of the host economy that influences subsidiary typeover and above the random host economy effect. The basic formulation was:T* ijk = α + β’X i + θ i μ i + γ’Y k + η k ν k + ζ ijkwhere X i is a vector of characteristics of firm “i”, Y k is a vector of characteristics of the hosteconomy “k”, θ i μ i represents the random firm effect, η k ν k represents the random host economyeffect, and ζ ijk represents an error term. The specific relationship modeled was:TYPE = a + b1 FIRMSIZE + b2 INTERNATIONAL + b3 NUMSUBS + b4 JAPAN +b5 EUROPE + b6 MARKET + b7 GROWTH + b8 DEVELOPMENT + b9OPENNESS + b10 TAXRATE + firm random effects + host country randomeffects22