About regression-discontinuity design (RDD)

242 views
Skip to first unread message

Lian Peng

unread,
Feb 22, 2018, 11:26:30 PM2/22/18
to pystatsmodels

I seached the stackoverflow and documentation of statsmodels, but I found little about RDD. For now, I found R packages named ‘rdrobust’ and ‘rdlocrand’ written by Calonico, Cattaneo, Farrell and Titiunik . So is there any implementation in python or statsmodels? Thanks in advance!

josef...@gmail.com

unread,
Feb 22, 2018, 11:45:20 PM2/22/18
to pystatsmodels


On Thu, Feb 22, 2018 at 11:12 PM, Lian Peng <lianpe...@gmail.com> wrote:

I seached the stackoverflow and documentation of statsmodels, but I found little about RDD. For now, I found R packages named ‘rdrobust’ and ‘rdlocrand’ written by Calonico, Cattaneo, Farrell and Titiunik . So is there any implementation in python or statsmodels? Thanks in advance!


statsmodels does not have anything specific to RDD yet. We have some of the building blocks like the models and patsy has the basis functions for splines.

I have seen some basic examples in some github repositories, but I never looked very closely nor systematically.

Essentially, I don't have an overview for this and only have some superficial knowledge about RDD. It is on the wishlist and we should eventually provide it, based on my current priorities, inverse probability weighting and difference-in-difference estimators is what I looked at more closely and to some extend at propensity score matching where I roughly know what code is available in Python. Even there we still don't have anything in statsmodels yet.

As aside: I had a long time problems with DiD because the simplest case is just testing an interaction term which can be done using the general setup and features of the models. It took me a while to figure out for which versions we can provide additional support. For RDD, I'm still at the stage of: just use an interaction term of spline and treatment threshold indicator.

Josef





josef...@gmail.com

unread,
Feb 23, 2018, 6:22:21 PM2/23/18
to pystatsmodels
Cattaneo https://sites.google.com/site/matiasdcattaneo/home looks like a good starting point with several articles, R, Stata packages and R Stata package articles, including data online.
I didn't check what they are doing for confidence intervals which are a difficult problem for non-parametric regression, but the abstracts sound good.

This sounds like a good project for inclusion in statsmodels, but might require quite a bit of work, unless we get permission to translate the R code.

Cattaneo was also a main reference for treatment effects in Stata which I used to implement IPW type analysis.

Josef

Reply all
Reply to author
Forward
0 new messages