fe

Question 1

How do we implement fixed effects? There are two methods - why are these equivalent? Is one method more desirable?

Answer 1

by including dummies, by demeaning each variable at the cluster level. Dummies are cluster intercepts and represent cluster means, demeaning does this same thing. Doesn’t matter which method’s used, points and SEs are the same.

Question 2

fe

Why would we use fixed effects?

Answer 2

if we think cluster invariant characteristics are biasing estimates

Question 3

fe

How can we test whether fixed effects are needed?

Answer 3

using a Hausman test.

Question 4

fe

What are some limitations of using FE?

Answer 4

we throw out a substantial amount of variation, reducing power. If there is measurement error and it’s relatively large relative to true values then estimates will be attenuated.

we assume the endogeneity is all associated with cluster invariance

can’t estimate cluster level coefficients

Question 5

how can we test if FE s valid?

Answer 5

strict exogeneity. toss in future x into model.

Question 6

how does measurement error in explanatory variables influence FE estimates?

Answer 6

attenuate point estimates. if there is high serial correlation over time (meaning all values are close to the mean) then what’s left, relative to measurement error, may be quite small, leading to large SE.