Finding ATE for different Research designs

Question 1

RCT, Survey Experiments ATE

Answer 1

E(Y1-Y0)

Question 2

Vignette Experiment, RCT Identification Assumptions

Answer 2

Treatment is independent of the potential outcomes

Question 3

SOO ATE

Answer 3

Need to meet the identification assumption in order to get the ATE from the observed outcomes by comparing the observed outcomes of the treated and untreated units within groups of X: E[Y|X, D =1] – E[Y | X, D=0]

Question 4

SOO Identification Assumptions

Answer 4

Selection on Observables, or that there is randomization between treatment and control within each stratum of X, and Common Support or that we observe participants and non-participants with the same characteristics

Question 5

SUTVA

Answer 5

Stable unit treatment value assumption, assumes that the potential outcomes for unit i are unaffected by unit j’s treatment assignment

Question 6

IV LATE

Answer 6

Because we can only measure the average causal effect of units whose treatment status is determined entirely on the instrument, we look at the compliers’ ate or the Local Average Treatment effect

Question 7

IV Identification assumptions

Answer 7

Exogeneity of the instrument (assignment of the insturment has to be randomly assigned), Exclusion restriction (the instrument only affects the outcome through the treatment), First stage relationship (have to show that being assigned to the instrument enables units to take up the treatment), Monotonicity (the instrument has a uniform effect on the treatment and rules out the existence of any defiers)

Question 8

Conjoint Experiments Identification Assumption

Answer 8

Treatment is randomly assigned, the addition of an extra survey item has no effect on how respondents evaluate the control items, and respondents don’t lie about their preference for a sensitive item

Question 9

Conjoint Experiments AMCE

Answer 9

Because you can’t identify the ATE since individual profiles arise rarely in the data, gotta use the average marginal component effect which identifies the causal effect of a particular value of the lth attribute of the profile j on the probability that the profile is chosen while holding all other attributes fixed

Question 10

RDD Identification Assumptions (sharp)

Answer 10

The potential outcomes of the units are continuous in the running variable (X) around the cutoff point (C) (basically the units that are close to each other on each side of the cutoff look very similar to each other), and that the units below the cutoff need to be abrupt (no gradual change)

Question 11

RDD LATE

Answer 11

Identifying the average treatment effects at cutoff, so it is the local average treatment effect, and need to create a bandwidth around the cutoff to measure the effect of the cutoff

Question 12

DiD Identification Assumptions

Answer 12

Parallel trends – had the treated units not have recieved the treatment, they would have followed the same trend as the control units. Also, any omitted variables realting to the treatment and outcome must be fixed over time

Question 13

DiD ATT

Answer 13

{Diff in means btwn T and C in post-period} – {Diff in means btwn T and C in the pre-period}