Test on Causal Inference Methods in Econometrics

Question 1: Is the fundamental problem of causal inference that for any given unit at a specific time, it is impossible to simultaneously observe both the outcome if treated and the outcome if untreated?

A. Ano

B. Ne

Explanation: The study materials define the fundamental problem as never observing both the outcome if treated ($Y_i(1)$) and the outcome if untreated ($Y_i(0)$) for the same unit at the same time. For treated units, only $Y_i(1)$ is observed, and for untreated units, only $Y_i(0)$ is observed.

Question 2: Which of the following is NOT a question to ask before trusting a Difference-in-Differences (DiD) estimate, according to the practical DiD checklist provided?

A. Are treated and control groups comparable?

B. Do pre-trends look parallel?

C. Is the dataset stored in a SQL database?

D. Could treated units anticipate the policy?

Explanation: The practical DiD checklist includes questions about group comparability, parallel pre-trends, and potential anticipation of the policy by treated units. It does not include a question about the type of database used to store the dataset.

Question 3: The Callaway-Sant'Anna (simple) and did2s (Gardner two-stage) estimators showed significant disagreement on the point estimate for the effect of opening a Vinmonopolet store.

A. Ano

B. Ne

Explanation: The study materials indicate that the Callaway-Sant'Anna and did2s estimators agreed on the point estimate to within sampling noise, meaning they did not significantly disagree.

Question 4: What is a key practical implication when the Two-Way Fixed Effects (TWFE) method places negative weights on individual unit-time treatment effects, as described in the study materials?

A. The TWFE estimate (τ̂) always represents a convex average of the average treatment effects on the treated (ATT).

B. A negative TWFE coefficient definitively proves that all individual treatment effects are also negative.

C. The sign of the TWFE coefficient (τ̂) can sometimes be opposite to the sign of every individual treatment effect.

D. A negative TWFE coefficient does not preclude the possibility that all actual treatment effects are positive.

Explanation: The study materials state, under 'Practical implication': 'A negative TWFE coefficient does not rule out that all genuine treatment effects are positive.' This directly aligns with the idea that a negative coefficient does not preclude the possibility of all actual effects being positive.

Question 5: The rel_year variable in the simulated EV panel data is calculated for all cities, including those designated as never-treated.

A. Ano

B. Ne

Explanation: The `rel_year` variable is defined in the R code as `if_else(first_treat > 0, year - first_treat, NA_integer_)`. This means that for cities where `first_treat` is 0 (indicating they are never-treated), the `rel_year` is explicitly set to `NA_integer_` rather than being calculated as `year - first_treat`.