Bookmark and Share

Notice: On March 31, it was announced that Statalist is moving from an email list to a forum. The old list will shut down on April 23, and its replacement, statalist.org is already up and running.


[Date Prev][Date Next][Thread Prev][Thread Next][Date Index][Thread Index]

re: Re: st: Regression Discontinuity Design using age + only binary variables


From   "Ariel Linden, DrPH" <ariel.linden@gmail.com>
To   <statalist@hsphsun2.harvard.edu>
Subject   re: Re: st: Regression Discontinuity Design using age + only binary variables
Date   Sat, 26 Nov 2011 12:08:54 -0500

A couple of notes: 

First, yes, the RD design is suitable to address your question, and in fact,
there are several studies that specifically use age as the continuous X
assignment variable (most notably for similar contexts that you are
reviewing, ie., receipt of Medicaid or Medicare services and the effect on
health services utilization, etc., or drunk driving fatalities around the
legal drinking age of 21). I can provide references upon request (unless
Cam, our Statalist reference source, supplies them as a follow-up) :-)

The fact that you have a binary outcome (and other binary covariates) does
not preclude the use of the design, or even the linear model, which is
addressed in the citations provided by Guo Xu). I would suggest you run
these models using both the parametric and non-parametric approaches and see
how well they concur. 

This brings up the second point: Austin Nichol's program -rd- performs the
"optimal bandwidth" selection that serves as the primary function in Imbens
software -rdob-(cited by Guo Xu below). Thus, you can do "one stop shopping"
with Austin's program. Also, to address my point of parametric vs
non-parametric approaches, Austin has, in fact, two program, -rd- which uses
parametric modeling and his original program - rd_obs- which use lpoly as
the basis for determining treatment effects. I'd run the data in both
programs and see if there is reasonable concordance. Also, don't forget to
run the balance checks on your covariates. I see many researchers skip this
step as if covariate balance is a forgone conclusion (which it is not)...

I hope this helps

Ariel


Date: Fri, 25 Nov 2011 10:18:37 +0000
From: Guo Xu <digitalepourpre@gmail.com>
Subject: Re: st: Regression Discontinuity Design using age + only binary
variables

Dear Richard,

In terms of reading, have a look at

Imbens and Lemieux (2007): RD designs: A guide to practice
van der Klaauw (2008): RD analysls: A survey of recent developments in
economics

and the chapter in Angrist and PIschke (2008): Mostly Harmless Econometrics

For Stata, there are several procedures, e.g.:
- - Nichols: http://ideas.repec.org/c/boc/bocode/s456888.html
- - Imbens: http://www.economics.harvard.edu/faculty/imbens/software_imbens

Hope that helps.

Guo


On 25 November 2011 08:12, Richardson <s.richardson@live.co.za> wrote:
> Hello,
>
> I'm analysing the treatment effect of a new policy which was introduced
> which entitled children under age 10 to receive free health care. I'm
> observing the effect this had on private health insurance subscription
> take-up in children under age 10 vs children over age 10. It would seem
that
> employing a regression discontinuity design would be appropriate given the
> cut-off point and jump in the data. All of the available data is binary
(0's
> or 1's) except for age.
>
> How would I estimate the treatment effect using an RDD assuming Sharp RD
> holds and then taking into account endogeneity and employing a Fuzzy RD
> approach?
>
> How can this be approached in Stata without overcomplicating the analysis.

*
*   For searches and help try:
*   http://www.stata.com/help.cgi?search
*   http://www.stata.com/support/statalist/faq
*   http://www.ats.ucla.edu/stat/stata/


© Copyright 1996–2014 StataCorp LP   |   Terms of use   |   Privacy   |   Contact us   |   Site index