tandard errors
From: "Duy Hinh Khieu" <[email protected]>
To: [email protected]
Date: Wed, 13 Jun 2007 07:44:05 -0400
X-Mailer: NetMail ModWeb Module
X-Sender: dhkhie2
MIME-Version: 1.0
Message-ID: <[email protected]>
Content-Type: text/plain; charset=3D"UTF-8"
Content-Transfer-Encoding: 8bit
X-MIME-Autoconverted: from quoted-printable to 8bit by hsphsun2.harvard.edu=
id l5DBiAM14169
Sender: [email protected]
Precedence: bulk
Reply-To: [email protected]
Errors-To: [email protected]
Thank you very much, Kit Baum, for your always helpful comments. I really w=
ant to thank all contributors to this forum for the unobservable, yet deepl=
y appreciated impact they create.
By the way, I have a propensity score matching question that I sent more th=
an twice and never get posted. I venture to attach the question here, hopin=
g to get it through (somehow, the email got screened out by its subject lin=
e?). I did not mean to violate the "scream" rules of the forum. I just thi=
nk my question has never got through.
Here it is:
I don=E2=80=99t know how to interpret certain parts of the output produced =
by pstest and can=E2=80=99t find any examples online to learn from. I am no=
vice at both pstest and stata. I need get the percentage of bias. The numbe=
rs from the attached output do not seem to be in %, and I don=E2=80=99t kno=
w how people reporting % get their number from if they use psmatch2 and pst=
est. How come the % reduction in bias from my output is greater than 3 digi=
ts, i.e., what does that number tell me and is it supposed to be less than =
100%? How do I know the numbers in the output are reasonable? I thought the=
p-values (in the t-test column) for the Matched rows should all be signifi=
cant because the sample and control groups are now matched and the variable=
s, similar, which makes the means no longer different and the tests of the =
means difference should be significant. I have two variables in the Matched=
row from the output that are not significant. What does that mean? Is ther=
e something wrong? If someone ca!
n attach their own output together with an interpretation for me to learn=
, I=E2=80=99d greatly appreciate. I am posting my output below.
. pstest ave3intcover ave3operincsale ave3totdebt ave3size
---------------------------------------------------------------------------=
-
| Mean %reduct | t-test
Variable Sample | Treated Control %bias |bias| | t p>|t|
------------------------+----------------------------------+---------------=
-
ave3intcover Unmatched | 6.864 12.138 -3.8 | -0.49 0.622
Matched | 6.864 9.1139 -1.6 57.3 | -1.75 0.080
| |
ave3operin~e Unmatched | .17478 .18209 -0.7 | -0.10 0.922
Matched | .17478 .19925 -2.5 -234.8 | -2.37 0.018
| |
ave3totdebt Unmatched | .34108 .33878 1.8 | 0.31 0.759
Matched | .34108 .33382 5.6 -215.1 | 0.74 0.457
| |
ave3size Unmatched | 8.026 7.7477 22.2 | 3.81 0.000
Matched | 8.026 7.9996 2.1 90.5 | 0.29 0.768
Hinh
-----Original Message-----
From: Kit Baum <[email protected]>
To: [email protected]
Date: Wed, 13 Jun 2007 06:38:43 -0400
Subject: st: Re: simultaneous equations with fixed effects and robust stand=
ard errors
No, your syntax is not correct. All endogenous variables and excluded
instruments must appear in one parenthesized expression in ivreg,
ivregress, xtivreg, ivreg2, xtivreg2:
xtivreg2 maturity (leverage my_binary_var=3Dsize profit) dividends
year86-year05, fe robust
If you want to put all 6 binary variables into the same equation, you
will obviously need more instruments. The equation as written above
is exactly identified, and you cannot do tests of overidentifying
restrictions.
Re the justification for this approach: if as I suggested you
searched the Statalist archives, you would find
http://www.stata.com/statalist/archive/2004-07/msg00710.html
You do not need to use probit or logit to generate consistent
estimates of the parameters of the equation as written above.
Kit Baum, Boston College Economics and DIW Berlin
http://ideas.repec.org/e/pba1.html
An Introduction to Modern Econometrics Using Stata:
http://www.stata-press.com/books/imeus.html
On Jun 13, 2007, at 2:33 AM, Duy Hinh Khieu wrote:
> Thank you - Kit Baum - very much for your comments. I am not really
> sure, however, of what you mean by standard IV estimators in the
> context of my model, which is mixed with one continuous endog var
> and 6 binary endog variables. If I use the xtivreg2 as follows, can
> you tell me if I translate your comments correctly into STATA code:
>
> xtivreg2 maturity (leverage=3Dsize profit) (my_binary_var=3Dsize
> profit ) dividends year86-year05, fe robust endog( leverage
> my_binary_var)
>
> I would need to do the bracket part of the code that contains
> my_binary_var six times because I have six binary endog variables
> and include them all in the above xtivreg2?
>
> The reason why I need to use probit or logit is not because I am
> concerned about a consistent estimator, but because I need to do a
> two-stage estimates and the first stage involves a binary variable.
> Am I right?
>
> If you have a source that states that the first stage estimation
> does not need to involve a probit or logit, I'd really appreciate
> your giving me that reference. All I have is Maddala (1983), pages
> 243-245.
*
* 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/
*
* 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/
