1. Chetty, Friedman, and Rocko use a teacher switching quasi-experiment to
study measures of teacher value-added.
(a) What were they trying to demonstrate with the teacher switching quasi-
(b) Suppose that teachers only switch between schools with very similar types
of students. What would that say about their results? Explain.
2. What do the ndings of Chetty et al. (Tennessee STAR paper) and Chetty,
Friedman, and Rocko (long-run eects paper) imply about the human capital
vs. signaling models of the returns to schooling?
3. In the lecture, we discussed a particular denition of what it means for teacher
ratings to be unbiased.” In particular, our denition was that ^ is a forecast
unbiased way of guessing if E(i j ^i) = ^i. (Equivalently, but using slightly
dierent notation, it is forecast unbiased if E(i j ^i = k) = k for all values of
(a) Suppose that there are exactly two types of basketball players, those who
make 75% of their shots and those who make 25% of their shots. Let i
be the fraction of shots that player i generally makes. I make a guess ^i
about a player i’s fraction of shots made by (i) observing one shot that
the player took, and then (ii) guessing :75 if the shot was made and :25 if
the shot was missed. Is this estimate forecast unbiased? Why not? (Hint:
Check if the forecast unbiasedness condition holds for each value of my
(b) Would it be any better if I had guessed 1 when the shot was made and 0
when it was not? Why not?
(c) In statistics and econometrics, a more common denition of unbiasedness
is that ^i is an unbiased way to estimate i if E(^i j i) = i. Translate
this denition into words. (Suggestion: You can refer to ^i as the guess
about person i’s value of theta” or the estimate of i’s value of theta.”)
(d) Consider the way I produced the estimate ^i in part (b). What is E(^i j
i = :75)? How about E(^i j i = :25)? (Hint: When Y is a dummy
variable, E(Y ) = Pr(Y = 1).) What do you conclude about whether or
not this estimator is unbiased in the sense described in part (c)?
4. We have seen evidence that teachers produce skills which are not captured by
contemporaneous test scores. Should we interpret those skills as being non-
cognitive, or do you think they might represent other kinds of cognitive skills
which are simply not measured by the exam?
5. In a couple of paragraphs, what are your thoughts on the use of value-added
models to evaluate teachers in schools?