search

gwb / rdd-coverage

1 stars 0 forks source link

Compare full Bayesian estimator to empirical Bayes (maximum marginal likelihood) #16

Open maximerischard opened 7 years ago

maximerischard commented 7 years ago

We think that it's important to do a full Bayesian treatment of the GP hyperparameters, because extrapolation is quite sensitive to those parameters (especially the credible interval). I think we need to:

  1. reference that claim
  2. demonstrate that claim in the RDD setting

It seems to me that we should just perform a set of simulations using the same settings as those explored by Zach, but using my empirical Bayes implementation of the 2GP estimate. We'd expect to see a reduction in the coverage.

maximerischard commented 7 years ago

Zach, could you send me your simulations code?

zjbranson commented 7 years ago

Yes, I'll do this soon today!

On Thursday, October 20, 2016, Maxime Rischard notifications@github.com wrote:

Zach, could you send me your simulations code?

— You are receiving this because you are subscribed to this thread. Reply to this email directly, view it on GitHub https://github.com/gwb/rdd-coverage/issues/16#issuecomment-255194239, or mute the thread https://github.com/notifications/unsubscribe-auth/AGeOVkfpgOefpJQr0-9hPl_qv5phdlgwks5q17ekgaJpZM4KceR7 .