When applying a simple linear regression model, why should caution be exercised in making predictions for values of x that fall outside the observed range?