Panel data-RE Flashcards
Difference between FE and RE?
FE-assume ui is correlated with xit
RE-ui is uncorrelated with xit
Model for RE?
yit=B0+B1xit+ui+eit
where vit=ui+eit
ui=between group error common to all observations within group (not indexed by t_
eit=within group error
what is ui?
bewteen-group error: common to all observations within group, not indexed by t
what is eit?
within-group error
True or false: eit and ui are distributed normally and iid
True
Is vit (composed of ui and eit) correlated with xit?
no, which means b1 will be unbiased and consistent
True or false: the composite error term vit is iid
False–The common error for observations in group i (ui ) results in correlation
between the composite error in period t (vit) and in period s (vis).
vit is not iid–what does this mean for OLS consistency and efficiency?
OLS is consistent but not efficient, and that traditional standard error formulas assuming i.i.d. errors are incorrect.
formula for intra-class correlation?
(sigma squared u)/(signa squared u+signa squared e)=
between group error variance/between group error variance+within group error variance
If FE assumptions hold (correlation between xit and ui), what does this mean for bias and consistency of RE model?
biased and inconsisent
If RE assumpttions hold (no correlation between xit and ui), what does this mean for consistency of RE and FE?
they are both consisent
If RE assumptions hold (no correlation between xit and ui), what does this mean for efficiency of FE?
FE model will be inefficient
If you get different answers from FE and RE, what does this mean?
you should use FE
How do you test if you should use FE or RE?
Hausman test
null and alternative hypotheses for Hausman test?
Null hypothesis: RE assumptions hold, both estimators consistent but RE is efficient.
Alternative: RE assumptions do not hold and the RE estimator is inconsistent
GLS transformation where var(u)=k σ^2u-how do you weight?
The GLS transformation divides the data by
Observations with greater variance get less weight.
Should you use RE or FE here: assignment to treatment was random at the school level, so we need not be concerned about correlation between treatment and the error term. However, the errors are not i.i.d.
RE
