Bookmark and Share

Notice: On April 23, 2014, Statalist moved from an email list to a forum, based at

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

Re: st: Re: comparing regression discontinuity treatment effects for different subsamples

From   John Antonakis <[email protected]>
To   [email protected]
Subject   Re: st: Re: comparing regression discontinuity treatment effects for different subsamples
Date   Tue, 12 Oct 2010 23:20:07 +0200

Hi Austin:

Thanks for this; I was not clear enough. In the following, suppose that selection is based on the following explicit rule (where cut-off is at the mean of the pretest):

group =1 if pretest of person i is less than or equal to mean of pretest
group =0 if pretest of person i is greater than pretest

We provide the treatment to group 1, and we estimate:

y = b0 + b1*(pretest - mean pretest) + b2*group + e

Here, b2 is the treatment effect and captures the jump in the discontinuity. Thus, constraining b2 to be equal across the two samples captures the difference in treatment effects across the two samples--or am I missing out on something?



Prof. John Antonakis, Associate Dean
Faculty of Business and Economics (HEC)
Department of Organizational Behavior
University of Lausanne
Internef #618
CH-1015 Lausanne-Dorigny

Tel ++41 (0)21 692-3438
Fax ++41 (0)21 692-3305

Home page:

On 12.10.2010 22:10, Austin Nichols wrote:
In that case, I strongly disagree with your advice--you are
constraining the slope of pretest to be the same on both sides of the
discontinuity, and not using any concept of a bandwidth around the
cutoff; local linear regression is the standard approach, not linear
regression.  Further, if you mean to subtract the mean of the
assignment variable when you say "pretest is mean-centered" then the
coefficient on group does not measure the jump in outcomes at the
cutoff unless the cutoff happens to be right at the mean of pretest.

On Tue, Oct 12, 2010 at 8:50 AM, John Antonakis<[email protected]>  wrote:
Hi Austin:

Using the "classical" RDD design, "group" is the treatment indicator;
pretest is the "cutoff" measure for assignment to group.

On 12.10.2010 03:34, Austin Nichols wrote:
John --
I don't understand your advice here at all--is group supposed to be a
treatment indicator?  Is pretest an assignment variable or a control
Prashant --
One can of course write a wrapper -program- containing several
estimators and -bootstrap- the whole thing, which then allows testing
across estimators--the -rd- package on SSC is no exception to that
general rule, but make sure you set the bandwidth exogenously if you
are using local linear regression as -rd- does.  Also -findit ivqte-
for one approach to quantile TE, and note that RD can be seen as a
version of IV; see refs cited in -help rd-.

On Mon, Oct 11, 2010 at 3:04 AM, John Antonakis<[email protected]>

You could use -suest-.  For example, suppose you have the following basic
specification (where pretest is mean-centered, to set the intercept to
cut-off value):

y = b0 + b1*pretest + b2*group + e

Estimate the model for each group, e.g.,

reg y pretest group if boys==1
est store boys
reg y pretest group if boys==0
est store girls
suest boys girls

Now you can do cross-equation tests, e.g.,

test [boys_mean]group = [girls_mean]group

Hope this helps.

*   For searches and help try:
*   For searches and help try:

© Copyright 1996–2018 StataCorp LLC   |   Terms of use   |   Privacy   |   Contact us   |   Site index