Appendix A: Figures and Tables
Figure 1. Chart of Variables and Relations
In this figure the dependent relations are modeled as they appear in the hypotheses in this research. The lines represent the relations between variables, and the arrows the influence on either the choice (the entry mode decision) or on the relation between a variable and the entry mode decision (a so-called moderating effect). The dependent variable in this research is Entry Mode, to which all the relations move.
-Figure 2. Integration-Responsiveness Grid: Strategic Focus and Organizational Adaptation (from Prahalad and Doz, 1987, page 25)
47
-Table 1. Cluster Analysis
Cluster Names Economies of Scale Global Competition Domestic Competition Product Differentiation N Global 3.21 3.49 1.99 2.45 87 Multidomestic 2.54 2.28 4.10 4.06 120 t-test 3.701 (0.000) 6.448 (0.000) -15.850 (0.000) -10.808 (0.000)
Table 2. Correlation Table Mean SD 1 2 3 4 5 6 7 8 9 10 11 12 13 Dependent Variable 1 Entry Mode .66 .47 Independent Variables 2 STRAT .42 .49 .076 3 INST 2.39 2.30 .055 -.070 4 INSTSTRAT .91 1.87 .081 .578** .501** 5 CULT 1.71 1.30 .017 -.004 -.462** -.162* 6 CULSTSTRAT .72 1.18 .061 .717** -.197** .238** .438** 7 SPEC 2.07 1.08 .077 .110 .049 .066 -.014 .110 8 SPECSTRAT .92 1.33 .058 .822** -.059 .475** .020 .615** .492** Control Variables 9 RESTR .29 .45 -.243** -.014 -.145* -0.62 .094 .022 .048 -.006 10 SIZE 3496 22094 .071 -.055 -.072 -.023 -.065 -.043 -.090 -.057 -.048 11 RELSIZE .88 6.51 .038 -.068 .076 -.037 -.092 -.057 -.009 -.060 -.045 -.013 12 INTEXP 18.12 26.57 .196** .173* .013 .080 .059 .241** .140 .193** -.071 -.011 -.261** 13 CEEENTRY .50 .50 .087 -.088 -.011 -.057 .114 -.042 .004 -.073 -.136 -.051 -.078 .204** 14 CEEEXP 4.01 3.67 .195** .013 .017 .057 .141* .016 .037 .056 .023 .020 -.098 .396** .371**
Table 3. Testing Results for Multicollinearity Variable Variance Inflation Factor (Tolerance) STRAT 7.326 (0.137) INST 2.985 (0.335) INSTSTRAT 3.442 (0.291) CULT 2.676 (0.374) CULSTSTRAT 4.651 (0.215) SPEC 2.193 (0.456) SPECSTRAT 6.865 (0.146) RESTR 1.123 (0.891) SIZE 1.104 (0.906) RELSIZE 1.326 (0.754) INTEXP 1.451 (0.689) CEEENTRY 1.161 (0.861) CEEEXP 1.455 (0.687)
Table 2 shows some statistics on the variables as specified for the binomial logistic regression in section 3.3. In general, for variables not to suffer from multicollinearity and influence the outcome of the statistical model, Tolerance should not be below 0.1 (or, similarly, the Variance Inflation Factor should not be above 10).
-Table 4. Binomial Logistic Regression Analysis
Strategic Focus Model Global Model Multidomestic Model
STRAT 3.154* (1.270) INST .036 (.137) -.003 (.135) .022 (.141) INSTSTRAT -.060 (.189) CULT .210 (.237) -.283 (.300) .182 (.251) CULSTSTRAT -.620† (.351) SPEC .655* (.276) -.186 (.275) .590* (.289) SPECSTRAT -.866* (.385) RESTR -1.137** (.430) -1.653* (.697) -.777 (.565) SIZE 0.000 (.000) .000 (.000) .000 (.000) RELSIZE -2.556*** (.766) -2.477† (1.468) -2.459** (.934) INTEXP -.335 (.407) .000 (.014) .034 (.026) CEEENTRY .056 (.063) -.019 (.693) -.736 (.553) CEEEXP .008 (.010) -.026 (.093) .114 (.097) C .040 (.999) 3.450* (1.509) -.222 (1.067) N 169 67 102 -2 Log Likelihood 177,800 67,079 103,116 Chi-square 38,254*** 10,899 32,568*** Nagelkerke R2 .281 .218 .372 % Correct 73.4 73.1 71.6
*** Correlation is significant at the 0.001 level (2-sided) ** Correlation is significant at the 0.01 level (2-sided) * Correlation is significant at the 0.05 level (2-sided) † Correlation is significant at the 0.1 level (2-sided)
Appendix B: Econometrical Background
Table 5. List of variables modeled
Variable Explanation
C (β0) Constant, the intercept of the function at the point where x is zero. An
alternative interpretation of the intercept for this research is the probability for full ownership for a firm with a multidomestic strategy, an institutional score of zero, a cultural distance of zero, no asset specificity, no restrictions to degree of ownership, a size of zero, a relative size of zero, no international experience in any country, no previous entry in CEE, and no previous experience in any CEE country.
STRAT The strategic focus of the firm captured in a dummy variable, where a 1 refers to a predominantly global focus and a 0 to a predominantly multidomestic focus.
INST The quality of the institutions of the host-country, ranging from -2.5 to 2.5, where a higher number refers to more advanced institutions.
INSTSTRAT The multiple of STRAT and INST, and therefore showing the INST scores for all firms that have a predominantly global focus.
CULT The cultural distance between the home and the host country, where a higher number refers to larger differences on the five dimensions as suggested by Hofstede (1980).
CULSTSTRAT The multiple of STRAT and CULT, and therefore showing the CULT scores for all firms that have a predominantly global focus.
SPEC The degree of asset specificity of the firm measured using the degree of R&D intensity, where a higher number means more relative investment in R&D and therefore more specific assets.
SPECSTRAT The multiple of STRAT and SPEC, and therefore showing the SPEC scores for all firms that have a predominantly global focus.
RESTR A dummy variable controlling for the existence of government restrictions on the degree of ownership, where a 1 refers to the existence of these and a 0 refers to the absence of such regulations.
SIZE The total size of the firm, measured in (millions of) total annual sales. RELSIZE The ratio of domestic to total sales, where a higher ratio implies a
(relatively) larger subsidiary.
INTEXP The number of countries the firm has international experience in.
CEEENTRY A dummy variable measuring whether the firm has set up a subsidiary in CEE prior to the current investment, where a 1 refers to a prior entry and a 0 refers to that no such entry has been made.
CEEEXP The number of countries in the CEE the firm has international experience in.
-Exhibit 1. Binomial Logistic Regression Explanation
There are many important differences between the use of ordinary least-squares (OLS) linear regression and a binomial logistic regression. For the OLS regression, five assumptions need to be fulfilled in order for the coefficients estimates to be the Best Linear Unbiased Estimators (BLUE) and have a number of desirable properties (Brooks, 2002). These assumptions are:
1. The error terms have a mean of 0
2. The variance of the error terms is constant and finite over all values of xt
3. The error terms are statistically independent of one another
4. There is no relationship between the error term and corresponding x 5. The error terms are normally distributed
These assumptions can be tested using specific statistical tests such as White’s test for heteroskedasticity (Assumption 2), Durbin-Watson’s test for autocorrelation (Assumption 3), or, alternatively, Breusch-Godfrey’s Serial Correlation LM test, and Jarque-Bera’s combined test of normality (Assumption 5).
However, apart from the problem that a linear regression is not suitable for probability testing as it allows for theoretically inadmissible results at the outer ends of the range (i.e. below 0 or above 1), the dataset used in this research is not able to meet al conditions for reliable and consistent outcomes as not all assumptions are met.