Statalist


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

Re: st: DiD estimation versus fixed effects when T>2


From   Nils Braakmann <nilsbraakmann@googlemail.com>
To   statalist@hsphsun2.harvard.edu
Subject   Re: st: DiD estimation versus fixed effects when T>2
Date   Thu, 5 Mar 2009 09:57:50 +0100

Dear Alison,

in your scenario you have several possibilities. Some remarks:

(1) Your treatment group dummy is (typically) fixed over time which
implies that it is indistinguishable from the unit fixed effect. If
your treatment group dummy varies within units (individuals/countries
etc.) over time, you might be in trouble as units may then select into
or out of treatment.
(2) You need interactions between the treatment group dummy and the
period dummy/dummies (see point 3). The coefficient gives the
divergence in trends in treatment and control group in the post
treatment period. Given that the assumption holds that both groups
would have experienced the same trend in the counterfactual situation
without treatment this coefficient is the treatment effect.
(3) There are two possibilities to set up the pre-/post-Treatment dummies:
(3.a) Define a single dummy that is "1" in all periods after treatment
and "0" before. You may also add additional time controls. The
treatment effect is then a time-weighted average of the effects in the
post-treatment years.
(3.b) Use year dummies and interactions between the treatment group
and each year. Say 2003 is the first year of treatment. The
treatment-group-year interactions for 2003, 2004... tell you how the
treatment effect evolves over time. Note, however, that the longer
time series make it somewhat more unlikely that the common-trend
assumption holds.
(3.c) If the treatment occurs in different years for different units,
you need to modify (3.b).

I am not exactly sure what you mean by "four different treatment
variables". I suspect you could use these to construct higher order
DiD-estimations (triple differences etc.) , but this is hard to tell
without knowing your exact situation. I am also not exactly sure what
your variable "treatment" is (in point (1) I suspected that this is
the treatment group dummy, but I'm not so sure now). Is this a
treatment group dummy or some sort of implicitly defined interaction
term (e..g. is this variable "1" only after a treatment occurred or is
it fixed over time)?

Hope this helps,
Nils

On Wed, Mar 4, 2009 at 4:00 PM, Riggieri, Alison C <ariggieri@gatech.edu> wrote:
> I am attempting to take panel data (5 years, for all 50 states +DC) and determine a DiD estimation.  I have four different treatment variables, all dummy variables.
>
> My question is this:  my adviser told me to to run "xtreg y x treatment 2002 2003 2004 2005, fe" to get a DiD estimation, but I'm not sure that is correct.  He told me to just add dummy variables to the fixed effects regression.
>
> Does that compute the DiD estimation or do I have to interact the treatment and the year?
>
> thanks in advance,
>
> Alison
>
> --
> PhD Student
> School of Public Policy
> Georgia Institute of Technology
> Atlanta, GA
> 508-410-1931
> *
> *   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/
>

*
*   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   |   What's new   |   Site index