Regression Assumptions
You test the assumptions by running tests and plots. However rejecting assumptions should not be
a goal itself. The regression analysis depends on a few assumptions such as normally distributed
error terms, linearity, homogeneity of variances, independence of error terms & no
multicollinearity. If an assumption is invalid, you may question how well your model serves as a
predictor of a phenomena.
Normally distributed error terms
(non)normally distributed error terms
● this is a problem since significance tests of coefficients are based on the assumptions of
normally distributed errors.
● assess through histogram, Q-Q plot
NOTE: on the exam, you should be able to interpret the the figures and state whether the
assumption is violated or not.
Homoscedasticity of variance
homogeneity of variance (homoscedasticity) or heteroscedasticity
● this is a problem (heteroscedasticity) because it makes it difficult to gauge the true
standard deviation of the error terms
● assess through scatter plot of predicted values (X) against residuals (Y)
1