ECN 388 Flashcards
What is MLR 1?
We believe the relationship to be linear in nature.
What is MLR 2?
We have a random sample of size n.
What is MLR 3?
The sample outcomes on all the x values are not the same.
What is MLR 4?
The error term, u, has an expected value of zero.
What is MLR 5?
The error term, u, has the same variance given any value of the explanatory variables.
Multicollinearity
When there is a strong correlation between x variables.
Three ways to detect multicollinearity:
- ) calculate correl coefficient and compare to R-squared.
- ) if p-values are greater than R-squared
- ) variance inflation factor
Variance Inflation Factor
VIF = 1/(1-r^2)
VIF > 10
Heteroskedasticity
Not a constant variance in the error term, u.
Violates MLR 5.
Four ways to detect Heteroskedasticity?
- ) Graphical (the error term against the predicted y)
- ) Goldfeld-Quant Test
- ) Park Test: multiplicative Heteroskedasticity
- ) Breusch-Pagan Test
Four ways to correct multicollinearity:
- ) add data
- ) fundamentally change the model
- ) drop a variable that contributes to multicollinearity
- ) do nothing and proceed with caution
Three ways to correct Heteroskedasticity:
- ) weighted least squares
- ) weighted least squares, different weight
- ) STATA fix: “robust”
What is Functional Form Misspecification?
Missing important variables that should be included.
How do we correct Functional Form Misspecification?
Add the needed variables
What are Proxy Variables?
When we know we need a certain variable in the regression but we can’t get it. Use a different variable as a proxy.
What is Measurement Error?
When the data that has been collected contains errors.
What is Missing Data?
Observations that have “holes” in them… Drop the observation.
What is Over-specification?
Adding variables to the regression that don’t belong there.
What is Omitted Variable Bias?
(AKA: under-specification): leaving out variables that should be in the model.
What is Endogeneity?
When an x variable is correlated with the error term.
Interpret a Log-Log model
On average, a one percent change in Xi results in a Beta % change in Y, ceteris paribus.
Interpret a Log-Lin model
On average, a one unit change in Xi results in a (Beta*100)% change in Y, ceteris paribus.
Interpret a Lin-Log model
On average, a one percent change in Xi results in a (Beta/100) unit change in Y, ceteris paribus.
