In order to evaluate the regression models statistically, the
following assumptions regarding the residuals (or standardised
residuals) were verified as according to Larsen et al. [13]:
1. The mean value of residuals (ej) is equal to zero.
2. The residuals can be plotted as a normal distribution.
3. The residuals have constant variance (homoscedasticity).
4. The residuals are independent (or random), i.e. there is no correlation
with regression coefficients or the response.
The first assumption is easily verified, while the remaining are
checked using plots of the standardised residuals. If assumption (2)
is correct, a normal probability plot of the standardized residuals
should produce a straight line. Scatter plots showing the relationship
between the standardised residuals and the predicted values
can be used to evaluate assumptions (3) and (4). If the residuals are
evenly and randomly distributed around zero, then the assumptions
hold true. These four points are verified in the next section.