tribution of the dots clusters less closely around the line, and becomes virtually
no correlation when the distribution approximates a circle or oval (the method
is ine¤ective for measuring a relationship that is not linear).
2.4.1 Pearson’s Correlation Coe‰cient
Consider the scatter diagram shown in Figure 2.14, where we have added a
vertical and a horizontal line through the pointðx;yÞand label the four quar-
ters as I, II, III, and IV. It can be seen that
In quartersIandIII,
ðxxÞðyyÞ> 0
so that for positive association, we have
X
ðxxÞðyyÞ> 0
Furthermore, this sum is large for stronger relationships because most of
the dots, being closely clustered around the line, are in these two quarters.
Similarly, in quarters II and IV,
ðxxÞðyyÞ< 0
leading to
Figure 2.14 Scatter diagram divided into quadrants.
COEFFICIENTS OF CORRELATION 85