Deck 7: Linear Regression Assumptions and Diagnostics

Question 1

Which one of these statements is not a Gauss-Markov assumption?

A) The error term has a conditional mean of 0
B) Influential observations are absent
C) The error term has constant variance
D) The errors are uncorrelated

Answer 1

B

Question 2

Why should we not include irrelevant variables in our regression analysis?

A) Your R-squared will become too high
B) Because of data limitations
C) It is bad academic fashion not to base your variables on sound theory
D) We increase the risk of producing false significant results

Answer 2

D

Question 3

How can we deal with the breach of the assumption about linearity?

A) Include a squared term
B) Include an interaction term
C) Use robust regression
D) Use the margins command

Answer 3

A

Question 4

What is the best way to find the exact top or bottom point of a squared effect?

A) Through derivation using values from the two coefficients
B) Excluding the squared term and predicting
C) Including the squared term and predicting
D) Graphing the results and comparing the top/bottom point with the value on the X-axis

Question 5

Name another way of modelling nonlinearity

A) Using the linktest command
B) Using interaction term
C) Using dummy variables
D) Using a bivariate regression model

Question 6

Which statistic(s) can help us detect multicollinearity?

A) Variance inflation factor (VIF)
B) F-statistic
C) Durbin-Watson
D) Tolerance values (1/VIF)

Question 7

What does heteroskedasticity mean?

A) The variance in the residuals is the same regardless of their predicted values
B) There is variance in the residuals
C) We are unable to produce residuals
D) The variance in the residuals differ depending on their predicted values

Question 8

What are the two ways we can check for heteroskedasticity?

A) We can examine a plot of predicted values vs the residuals
B) We can run the Hausman test
C) We can run the hettest command
D) We can compare the F-test of two models

Question 9

What does robust regression do?

A) Performs an OLS regression with more trustworthy standard errors
B) It gives a weight to each unit based on their distance from the mean of Y
C) Performs three types of regression analysis and presents the mean results
D) It gives a weight to each unit based on their total influence on the model

Question 10

Which one is not a measure of influential (or potentially influential) observations?

A) Leverage
B) Cook-Weisberg
C) DFBETA
D) Cook's distance