In their summary of the assumptions of multiple
regression, the first of four assumptions given focus by
Osborne and Waters (2002) is the normality
assumption. Osborne and Waters state: “Regression
assumes that variables have normal distributions” (p.
1). They do not explicate which variables in particular
they are referring to, but the implication seems to be
that multiple regression requires that the predictor
and/or response variables be normally distributed. In
reality, only the assumption of normally distributed
errors is relevant to multiple regression: Specifically, we
may assume that errors are normally distributed for any
combination of values on the predictor variables.