click below
click below
Normal Size Small Size show me how
Regression
Violations of Classic Assumptions
Question | Answer |
---|---|
Which classical assumption(s) does an Omitted Variable violate? | Violates Classical Assumptions I and III |
What are the consequences of an Omitted Variable? | 1) Estimated coefficient doesn’t equal actual coefficient. 2) Bias is forced onto another coefficient, causing the estimated value of the coefficient to change. 3) Increases by decreasing variance. |
What factors may indicate Omitted Variable bias? | 1) Unexpected signs on the coefficient. 2) Too big of coefficients (if positive bias on a positive coefficient). |
How do you solve the problem of Omitted Variable bias? | Add omitted variable or proxy variable. |
Which classical assumption(s) does a Redundant Variable violate? | Violates Classical Assumption VI |
What are the consequences of a Redundant Variable? | 1) Does NOT introduce bias. 2) Increases variance. |
How do you detect a Redundant Variable bias? | 1) Decreased R2. 2) Wald Test. |
How do you solve the problem of Redundant Variable bias? | Drop irrelevant variable |
How do you use to decide whether to include Omitted or Redundant Variables? | 1) Theory. 2) t-tests. 3) Adjusted R2 decreases if the improvement in overall fit due to addition of the independent variables to the regression does NOT outweigh loss in degrees of freedom. 4) Bias. |
Which classical assumption(s) does omitting an Intercept violate? | Violates Classical Assumption II: the error term has a zero population mean. Intercept usually absorbs error. |
Which classical assumption(s) does Multicollinearity violate? | Violates Classical Assumption VI |
What are the consequences of Multicollinearity? | 1) Does not cause bias 2) MIGHT CAUSE WRONG SIGN DUE TO INCREASED VARIANCE 3) Increases variance because it impacts r sub1,2 4) T-scores Fall because SE increases 5) Will not fall much |
How do you detect Multicollinearity? | 1) High Adjusted R2 and low t-scores. 2) High simple correlation coefficients (r sub1,2). 3) High Variance Inflation Factor (VIF > 5). |
How do you interpret VIF? | When VIF = 5, this means you have 5 times the variance in the model than you would have without the multicollinearity. |
How do you solve the problem of Multicollinearity? | 1) Do nothing 2) Drop variables 3) Transform variables 4) Increase sample size |
Which classical assumption(s) does Serial Correlation violate? | Violation of Classical Assumption IV |
What are the consequences of Serial Correlation? | 1) No bias in coefficient estimates. 2) TRUE Increased variance in coefficient estimates. 3) Distorts SEE portion of SE. 4) OLS no longer BLUE. 5) OLS underestimates standard error of the coefficients. |
How do you detect Serial Correlation? | 1) T-scores appear larger than they really are, leading to Type I error More likely to make a Type I error (reject a null hypothesis that is true) 2) Durbin-Watson test |
How do you solve the problem of Serial Correlation? | 1) Better Specification 2) Generalized Least Squares |
Which classical assumption(s) does Heteroskedacity violate? | Violation of Classical Assumption V Most common with cross-sectional models Comparing proportionately different instances results in inconstant variance in error term |
What are the consequences of Heteroskedacity? | 1) No bias in coefficient estimates, t-scores are larger. 2) TRUE Increased variance in coefficient estimates 3) Distorts SEE portion of SE 4) OLS is no longer BLUE 5) OLS underestimates standard error of the coefficients. |
How do you detect Heteroskedacity? | 1) Plot residuals to see if variance is constant. If bell or flower shape emerges, there is heteroskedasticity. 2) Park Test (for proportionality) 3) White Test |
How do you solve the problem of Heteroskedacity? | 1) Weighted Least Squares. 2) Redefinition of Variables. 3) Heteroskedasticity Corrected Standard Errors. |