CK020 - Linear Regression

Question 1

What are the assumptions of a linear regression model?

Answer 1

• Error terms are homoscedastic
• Error terms are uncorrelated
• Error terms are normally distributed
• Linearity
• No (multi)colinearity

Question 2

What are ‘standardized residuals’ ?

Answer 2

Residuals where the heteroscedasticity is removed (so they are homoscedastric residuals)

Question 3

What are ‘standardized residuals’ used for?

Answer 3

• Assessing homoscedasticity
• assessing normality of the residuals

Question 4

What are ‘studentized residuals’ ?

Answer 4

Residuals with a known (t-) distribution

Question 5

What are ‘studentized residuals’ used for?

Answer 5

Identifying outliers

Question 6

How to check the ‘normality assumption’ ?

Answer 6

Plot standardized residuals (y) against the theoretical residuals (x) and check if this is a linear line.

Question 7

What is ‘homoscedasticity’ ?

Answer 7

Standardized/studentized residuals are randomly spread around 0 with constant variance

Question 8

How to check the ‘homoscedasticity assumption’ ?

Answer 8

By plotting the standardized/studentized residuals (y) against the fitted values/covariates (x)

Question 9

What to do with ‘heteroscedasticity’ ?

Answer 9

• Variable transformation (not recommended)
• Weighted Least Squares

Question 10

What is an ‘outlier’ ?

Answer 10

Any observation that does not ‘fit the model’

Question 11

What is a ‘high leverage point’ ?

Answer 11

Observation with extreme predictor value(s)

Question 12

What are ‘influential values’ ?

Answer 12

Observations that have excessive influence on the model

Question 13

Why should you do model diagnostics?

Answer 13

To evaluate the assumptions are reasonable for the data at hand