Part 4: Questions and Answers
Chapter 19
Q: Do we ever observe the ATE? Or are we just trying to estimate ATE?
A: We never observe individual treatment effects, and ATE is their average. When we have random assignment or, close to partialling out endogenous variation in x, we have a chance to estimate ATE. But in the data itself, we never observe it.
Chapter 23
Q: In a panel regression, when we are taking the differences of log values on either side, am I thinking right that we are supposed to report percentage point changes instead of percentage changes?
A: No, it is actually percent changes. Log difference is an approximation of relative difference in percent. We shall use percentage points when the x or y variable is already in percent, such as share of immunization for children (Ch23, case study B). Then, we did not take logs, but kept percent, so difference between percent values is the percentage point.
Q: With unbalanced panel data, what should I do if I see that there is a great deal of missing values for some years in particular
A: If you see that, say, for a panel data on countries for the 1992-2018 period, data is missing for many countries till 1995, you may drop those years, rather than keep them empty. In this case, you may see that the estimated average will more reflect countries with full time series. If having shorter series is correlated with some key variable, this can bias the estimate.