Collinearity
is usually no problem for predictive purposes. The correct answer
is (c), correlated variables may well be combined to obtain
a more stable measurement of the underlying biological phenomenon.
An example is taking the mean of diastolic and systolic blood
pressure. Related variables may also be combined in a summary
variable. For example, an atherosclerotic risk variable may
be created based on the presence of various signs that indicate
atherosclerosis (Krijnen
et al., 1998). Also, we might count the number of signs
to form such a variable (Harrell
et al., 1985).