Week 8: Chapter 14 Flashcards
What does the first difference model do?
Transformation that removes time fixed effects ai
What does the Fixed Effects model do?
Transformation that removes time fixed effects ai
What does the Random Effects model do?
when ai is uncorrelated with the xitj (typically estimated by GLS)
Show the steps of obtaining the FE estimator
look at notes
What is unique about the dof in the FE model? What happens to the dof’s
- For every observation i, the FE model loses a dof due to time demeaning
- N(T-1) - k (correct way to find DOF in FE model)
When T=2, do we use FD or FE?
The estimates will be identical. FE needs to include a dummy variable for the second time period to be identical to FD, because FD includes an intercept
For large N and small T, FE/FD?
- choose FE or FD based on the relative efficiency of the estimators
What determines efficiency?
Serial Correlation: Corr (us, ut) does not = 0 for some t do not = s
Is FE or FD used more often?
When Uit are serially uncorrelated, FE is more efficient than FD
If strict exogeneity is violated: FE/FD?
then use FE over FD - it has less bias
Units appearing only once don’t get used in fixed effects analysis, WHY?
Time-demeaning for these individuals yields zeros for all their variables
What is the formula for vit being serially correlated in the RE model?
σ^2a / σ^2a + σ^2u, s does not = t
Define the formula for quasi-demeaned data
look at notes
Write formula for RE model
yit - θȳi = 𝛽0(1 -θ)+ 𝛽1(xit1 - θx̄i1)+…+ 𝛽k(xitk - θx̄ik)+(vit - θvi)
What does the RE model do?
Subtracts a fraction of the time average from xitj and yit
