DD & DDD

Question 1

What do we use as the counterfactual?

Answer 1

time trend for observations that belong to the control group

Question 2

1st DD assumption

Answer 2

E[Y_0igt | g,t] =

Question 3

what assumption do we check in the pre-period ?

Answer 3

common/parallel trends assumption:

want to observe common/parallel trends in treatment vs control groups in the pre-period

Question 4

What are the 3 SE problems ?

Question 5

how do you correct group correlation?

-no longer I.I.D, everyone in group g has a correlated

Answer 5

the moulton factor

MF= sqrt(B₁ hat)/sqrt(B₂ hat)

Question 6

How do you correct for the SE problem of serial correlation?

-observations for the same underlying unit are not independent over time

Answer 6

cluster SE approach, run regressions at group level

Question 7

How do you correct for the SE problem of heteroskedasticity?

-variability of a variable is not equal across the range of values of a second variable that is predicts

Answer 7

robust standard error –> always use the bigger of the robust SE and SE

Question 8

IV

what is the word meaning of exclusion restriction?

Answer 8

the instruments Z_i are not correlated with any other determinants of the dependent variables

Question 9

IV

what kind of treatment effect does IV create?

Answer 9

LATE

–> local refers to the part of the sample that is influenced by the instruments

Question 10

IV/2SLS

when using IV/2SLS what happens to the SE?

Answer 10

IV/2SLS almost always increases SE on coefficient for variable of interest

Question 11

IV w/ heterogenous potential outcomes

2 assumptions:

Answer 11

1. Z is binary
2. assume a constant causal effect

Question 12

IV w/ heterogenous potential outcomes

internal validity?

external validity?

Answer 12

internal validity: unbiased estimator for some sample

external validity: predictice value of studys findings (in other sample)w

Question 13

IV w/ heterogenous potential outcomes

four assumptions that give internally valid LATE?

Answer 13

1. non zero 1st stage
2. exclusion restriction
3. monotonicity assumption
4. treatment is as good as randomly assigned

Question 14

IV w/ heterogenous treatment effects

definition of complier, never taker, always taker

Answer 14

complier: people who change their behavior based on the instrument

always taker: people who always take treatment regardless of the instrument

never taker: people who never take treatment regardless of the instrument

Question 15

IV w/ heterogenous treatment effects

is LATE usually equal to the effect of the average treatment effect on the treated?

Answer 15

no

since IV is based on the compliers

Question 16

IV

what is the Bias of IV approach?

Answer 16

1. “weak” instrument vs OLS, b/c a small amount of “bad” variation leads to more bias in IV than OLS

–> weak instrument if F-stat for 1st stage is <10

2. can never really know if exclusion restriction is ok

3.

Question 17

Kernal

difference between LOWESS and kernal regression

Answer 17

1. lowess regression calculates the estimated CEF points at X* by using a locally weight regression of yi’s near x*, rather than a locally weighted mean of yi’s near x*
2. the lowess regression uses a variable bandwidth(hi) that increases where the data is sparse and decreases where there are many xi observations