This question is a general question not related to any particular data. What happens to the errors when the regression line is fit on curving data? (Remember the errors are supposed to be normally distributed, have constant variance and not have a pattern, i.e., be independent.)