01.11.2012 Views

Homework 2

Homework 2

Homework 2

SHOW MORE
SHOW LESS

You also want an ePaper? Increase the reach of your titles

YUMPU automatically turns print PDFs into web optimized ePapers that Google loves.

c) Is there a discontinuous jump in the distribution of x2 at the RDD threshold ¯x1 = 0.2?<br />

How could one implement a test for this?<br />

d) Now assume that the treatment effect α is random with E(α) = 3 and the distribution of<br />

α is independent of (x1, x2). Show that the conditions for a sharp RDD are satisfied here<br />

as well, i.e. the RDD estimator identifies α. Consider the assumptions put forward in Van<br />

der Klaauw (2002) and check formally that they are satisfied in this case. Motivate the<br />

identification result. What is the control function k(S) in this context? Be as specific as<br />

possible.<br />

Remark: This is a theoretical exercise. To solve this problem, it is not necessary to implement a<br />

TSP program for estimation purposes.<br />

[ 7 credits ]<br />

5.) (Fuzzy RDD: Angrist and Lavy 1999, Maimonides Rule, PC Pool Problem Set 3, Problem 4)<br />

Use the programs and the data provided as part of the third PC Pool Problem Set. For the<br />

problem analyze only math scores as outcome variables for the fourth grade.<br />

a) Analyze whether the variable ’percent disadvantaged’ shows discontinuous jumps at the<br />

tresholds used to split classes. What do these results imply regarding the question as to<br />

whether it is necessary to control for the variable ’percent disadvantaged’ when estimating<br />

the RDD estimate of the treatment effect of class size?<br />

b) Implement the 2SLS estimate of the effect of class size on math scores in the fourth grade<br />

controlling for enrollment using an appropriate polynomial specification of enrollemnt and<br />

percent disadvantaged as control variables. Discuss the results.<br />

c) Implement the local Wald estimator for the RDD based on a local linear regression (using<br />

a rectangular kernel) of class size on expected class size on both sides of each threshold.<br />

Use these nonparametric estimates from the first stage to estimate in a second stage the<br />

fuzzy RDD estimate of the effect of class size<br />

E(Yi|<br />

ˆρs = lim<br />

∆→0<br />

¯ Ss < S < ¯ Ss + ∆) − E(Yi| ¯ Ss − ∆ < S < ¯ Ss)<br />

E(Di| ¯ Ss < S < ¯ Ss + ∆) − E(Di| ¯ Ss − ∆ < S < ¯ Ss)<br />

for the s th threshold ¯ Ss (using the notation in the lecture). Discuss the differences in<br />

methods and results in comparison to part b).<br />

Remark: The solutions to the problem should include the TSP–programs, which you wrote to<br />

solve the problems, together with the output on paper. The discussion of the results should make<br />

reference to the computer output.<br />

5<br />

[ 6 credits ]

Hooray! Your file is uploaded and ready to be published.

Saved successfully!

Ooh no, something went wrong!