Statalist


[Date Prev][Date Next][Thread Prev][Thread Next][Date index][Thread index]

Re: st: Difference-in-difference regression models


From   "riti@ucla.edu" <riti@ucla.edu>
To   statalist@hsphsun2.harvard.edu
Subject   Re: st: Difference-in-difference regression models
Date   Wed, 24 Oct 2007 10:21:55 -0700

Thanks, Austin. Sorry to all that this was the wrong outlet for the question. I'm writing on behalf of a team working on a project for which we are using difference-in-difference estimates for the first time -- we all are feeling terrible that it appeared that we are misusing this forum. Thanks again for the direction.

Quoting Austin Nichols <austinnichols@gmail.com>:


Riti assures me this is not a homework question.  I still think the
UCLA web book is a good place to start; I am not always clear on where
understanding about interpreting the coefs on indicator vars breaks
down...
but here goes:

clear
input score   bonus control baseline
      0.4     1     0       1
      0.2     1     0       1
      0.1     1     0       0
      0.9     1     0       0
      0.5     0     1       1
      0.6     0     1       1
      0.7     0     1       0
      0.76    0     1       0
end
g bonusbaseline=bonus*baseline
g controlbaseline=control*baseline
reg sc bonus bonusb baseline, nohe
reg sc bonus bonusb controlb, nohe

Q: What explains the difference in the
coefficients for bonusbaseline across
regression models?

A: The coef on bonusbaseline measures the increment
due to bonusbaseline, given other coefs.

In the first regression, that's
(mean over obs w/only X2=1)-[
(_cons=mean where all three X=0)+
(mean over only X1=1 less _cons)+
(mean over only X3=1 less _cons)
]=
.3-(.73+(.5-.73)+(.55-.73))=-0.02

Similarly, in the second regression,
the coef on bonusbaseline measures the
diff in means across group 1 defined
by bonus and baseline equal to 1 (obs
1 and 2, mean .3) vs group2 defined by
bonus=1 and baseline equal to 0 (obs 3
and 4, mean .5) which is -0.2

This is all probably easier to see if you:

collapse sc, by(control bonus baseline)
g bonusbaseline=bonus*baseline
g controlbaseline=control*baseline
li, noo clean
reg sc bonus bonusb baseline, nohe
reg sc bonus bonusb controlb, nohe

but no doubt someone else can give a
more accessible interpretation of regression
coefficients.



*
*   For searches and help try:
*   http://www.stata.com/support/faqs/res/findit.html
*   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