Two-stage Least Squares (2SLS)
You have cross-sectional data on individuals’ log earnings Y
i
, years of schooling S
i
, and an instrument Z
i
equal to the (demeaned) distance to the nearest college. Assume Y
i
= α + βS
i
+ u
i
, E[u
i
| Z
i
] = 0, and that Z
i
affects Y
i
only through its impact on S
i
(think about the usual regression here, what assumptions should you make regarding the error of this second regression?)
Write down the first-stage and reduced-form regressions. Briefly explain what each coefficient measures.
Show how to compute the 2SLS estimate of β as a function of the first-stage and reduced-form coefficients.
Describe how you would test the instrument for relevance in the first stage, and state a rule-of-thumb threshold for the F-statistic.
Suppose the first-stage F-statistic is 5. What concerns does this raise for your 2SLS estimates?
Interpret the parameter identified by 2SLS in this context. Which subpopulation does it refer to?
Leave a Reply