1. You may have figured out that we can think of the method of randomization as an example of using the instrumental variables technique. Let us consider an example. Suppose we want to know the effect of missing classes on a student's final score.
2. (a) If we just ran an OLS regression of final score on having missed classes, what would be some confounders? Would we be able to control for all of them?
3. (b) Now suppose we did the following randomization (which would be a terrible idea to do in practice): for each student, we flip a coin. If the coin lands heads, the student has to attend all classes. If the coin lands tails, the student may only attend 50% of the classes. Can we now use the coin toss resulting in heads as an instrument for attending classes? Would it be a relevant instrument? Would it be a valid instrument?