Page 118 - Culture Technology Communication

National Level Culture and Global Diffusion 103

demonstrate the high multicollinearity between variables. For ex-
ample, GDP, an economic variable, is highly correlated with two in-
frastructure variables, the number of fax machines (.914, p .01) and
teledensity (.885, p .01). The gender empowerment measure is
highly correlated with teledensity (.689, p .01) and PCs per thou-
sand (.692, p .01). Thus, the empirical analysis will be performed
under the limitation of multicollinearity and the number of coun-
tries reporting data.
Before continuing with the analysis it is interesting to look at
the explanatory power of a few variables from each category to see
if one group dominates the others. From Table 3 it can be seen that
the cultural variables are slightly less powerful predictors of start
time than the economic and infrastructure variables. This was ex-
pected conceptually and supports previous diffusion research that
has found economic factors to be the strongest predictors of adoption.
The position of this research is not to show that cultural variables
are the strongest predictors but that they can help in explaining
adoption and this position is supported by this first step in the
analysis. 13
Due to the constraints of multicollinearity and data availabil-
ity, two methods for creating viable models (or combinations of IVs)
were used. The first method identifies a dominant independent
variable, with subsequent independent variables (IVs) being added
based on their relationship with the dominant IV. This first
method sacrifices comparability between models for including a
wider range of variables and potentially a larger number of coun-
tries in the analysis.
The results of this method are models that explain a compara-
tively larger amount of variance in the dependent variable, however
the models are not directly comparable with one another because
each combination of independent variables creates a unique sample
of countries reporting data for those particular variables. This prob-
lem is addressed by the second method. The second method uses a
sub-sample of fifty-five countries, all of which report data for eight of
the strongest predictors. This method allows comparisons across
models to be made.

INTER-COUNTRY DIFFERENCES

The results of methods 1 and 2 are given below; the correlation
matrices for both the full sample and the sub-sample of countries, as
well as a list of the countries in the sub-sample, are provided in