I am doing fuzzy RDD recently and I am facing some challenging moment dealing with Stata. I am using the command -rdplot- and -rdrobust-. I have some questions regarding the fuzzy RDD and the commands;

  1. When the cut-offs is not known, is it still possible to seek for a discontinuity using fuzzy RDD (or the sharp one)?

  2. the -rdrobust- command has an option fuzzy(treatment) to implement the fuzzy RDD, it has results on the coefficient. Let say my outcome variable is LN number of passengers, running variable is a distance between two cities. When I run the fuzzy RDD using a specific cut-off, I obtained significant result with the value of coefficient is 7.833. What does it mean? I read a paper by Imbens and Lemieux (2007) as "the ratio of the jump in the regression of the outcome on the covariate to the jump in the regression of the treatment indicator on the covariate". So, how do I interpret the 7.833?

Thank you.

  • $\begingroup$ On (2), see this question about fuzzy RD as IV, a kind of Wald estimator. As far as interpretation, you need to say more about what the treatment is. $\endgroup$ – Dimitriy V. Masterov Jul 27 '16 at 1:48
  • $\begingroup$ Thank you. So, the number can not be interpreted directly? It is more useful if I claim only the sign of the coefficient? $\endgroup$ – putut purwandono Jul 28 '16 at 2:48
  • $\begingroup$ @putut_purwandono If LN is natural log, then the effect of treatment is a semi-elasticity which corresponds to an astounding $100 \cdot (\exp{7.833} - 1) = 252,148.55\%$ increase in passengers. $\endgroup$ – Dimitriy V. Masterov Jul 28 '16 at 6:10

On (1), take a look at these 2-3 papers that use a data-driven approach:

  • Chay, K. Y., P. J. McEwan, and M. Urquiola (2005). The central role of noise in evaluating interventions that use test scores to rank schools. American Economic Review 95(4), 1237– 1258
  • Bertrand, M., R. Hanna, and S. Mullainathan (2010). Affirmative action in education: Evidence from engineering college admissions in India. Journal of Public Economics 94 (1), 16–29.

They essentially run a series of regressions of treatment on a dummy that equals 1 after each possible cutoff point and choose the one cut-off that gives the highest $R^2$ of the regression.

My co-author Matt Backus and his student Sida Peng have a working paper on Identification and Estimation of Discontinuities that uses some machine learning methods to do this, but there is no public draft yet.

On (2), see this question about fuzzy RD as IV, a kind of Wald estimator.

| cite | improve this answer | |
  • $\begingroup$ @pututpurwandono Did this clear things up? $\endgroup$ – Dimitriy V. Masterov Aug 9 '16 at 10:20

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service, privacy policy and cookie policy

Not the answer you're looking for? Browse other questions tagged or ask your own question.