Preparation

This week’s session is concerned with…

As a preparation,

1. Watch the videos on:

   • the basics of the Roy model (slides)

   • the marginal treatment effect (slides)

   • extrapolation and control functions (slides)

   • looking at the binary case graphically (slides)

2. Read sections 3.1 to 3.3 of Heckman and Vytlacil (2007).

3. Read sections 1 and 2 of Mogstad, Santos, and Torgovitsky (2018). Fear not, this only relates to the last of the learning objectives under “Methods”.

4. Skim very briefly over Kline and Walters (2019). You should only remember the structure, don’t pay attention to details. You may, however, find Figures 1 and 2 helpful in case you did not find the last screencast all too clear.

5. Consider the notebook creating the graphs from the last video. Make sure you understand what the figures represent

   1. What are “observed” data points, i.e., which values can you calculate as moments of an infinite amount of data?

   2. What is structure that we only know because we set the data generating process in this exercise? (think of it as a Monte Carlo study without actually generating data from the data generating process — usually we do these studies in order to understand the finite sample properties of estimators; here it is purely about identification)

   3. What can we calculate in an internally consistent way (i.e., without assuming anything but standard exogeneity/exclusion conditions on \(Z\) and additive separability / uniformity)?

   4. Where do we need to extrapolate in order to make predictions, i.e., make assumptions about quantities we cannot calculate despite having access to an infinite amount of data generated from such a model?

6. Take the quiz on eCampus