Page 48 - Cultures and Organizations

Studying Cultural Differences 33

A correlation coefficient is said to be (statistically) significant if it is

sufficiently different from zero (to the positive or to the negative side) to
rule out the possibility that the similarity between the two measures is
due to pure chance. The signifi cance level, usually 0.05, 0.01, or 0.001, is
the remaining risk that the similarity is still accidental. If the signifi cance
level is 0.05, the odds against an association by chance are 19 to one; if it
is 0.001, the odds are 999 to one. 7
If the correlation coefficient between two variables is 1.00 or 1.00, we

can obviously completely predict one if we know the other. If their correla-
tion coeffi cient is 0.90, we can predict 81 percent of the differences (called
the variance) in one if we know the other; if it is 0.80, we can predict 64
percent of the variance, and so on. The predictive power decreases with
the square of the correlation coefficient. If we have a lot of data, a correla-

tion coeffi cient of 0.40 may still be signifi cant, although the fi rst variable
predicts only 0.40 0.40 16 percent of the variance in the second. The
reason we are interested in such relatively weak correlations is that often,
phenomena in the social world are the result of many factors working at
the same time: they are multicausal. Correlation analysis helps us to isolate
possible causes.
In the case of three or more measures, we can choose one as our depen-
dent variable and calculate the combined effect of the remaining, independent
variables on this dependent variable. For example, we could measure not
only the height but also the shoulder width of our hundred randomly picked
test persons, and these two “independent” variables together would correlate
with our “dependent” weight measure even more strongly than height alone.
A statistical technique called regression allows us to measure the contribution
of each of the independent variables separately. In our analysis we often use
stepwise regression, a method to sort the independent variables step-by-step
in order of their contribution to the dependent variable. This contribution is

usually expressed as a percent of the variance in the independent variable.
In a stepwise regression of the body measures of our imaginary hundred

persons, we might find, for example, that height contributed 64 percent to the
variance in weight, and height plus shoulder width contributed 83 percent.

For readability reasons, correlation coefficients and regression results
in this book are given in the endnotes; the text refers to the conclusions
drawn from them and sometimes to percentages of variance explained.
Readers interested in additional statistical proof are referred to Geert’s
book Culture’s Consequences, 2001.