st: RE: re: std error covariances between equations

 From "FEIVESON, ALAN H. (AL) (JSC-SD) (NASA)" To "'statalist@hsphsun2.harvard.edu'" Subject st: RE: re: std error covariances between equations Date Tue, 17 Sep 2002 08:59:10 -0500

```Kit - Oops!  I am guilty of not reading the manual. I was unaware of the
-stddp- command. Thanks, Kit for pointing it out. As you say, it should be a
relatively simple manner to get the cross-covariance terms from the variance
of the difference. But I still question of what use the variance of a
difference between two equations really is. In many applications, the
quantities estmated by separate equations are not comparable (different
units, discrete-continuous, etc). Wouldn't it be more useful to have a
direct command to obtqin the cross-covariance? Perhaps we could get a
response from someone at Stata Corp. on this.

Al Feiveson

-----Original Message-----
From: Kit Baum, Faculty Micro Resource Center [mailto:fmrc@bc.edu]
Sent: Friday, September 13, 2002 9:04 PM
To: StataList
Subject: st: re: std error covariances between equations

I believe that Alan's problem, in a two-equation context, could be viewed as
looking for the covariance term
cov [x1 0] V [0 x2'] = x1 V12 x2'
using V12 to refer to the NE block of the
covariance matrix. Alternatively, we could be looking for
cov [0 x2] V [x1' 0] = x2 V21 x1'
where V21 is the SW block of the covariance
matrix. Since x1 and x2 may have different numbers of columns, these
off-diagonal blocks are not symmetric.

The predict option stddp calculates the standard error of the difference in
linear predictions. Using the same algebra, and the formula given in [r]
predict,
this is x1 V11 x1' - x2 V21 x1' - x1 V12 x2' + x2 V22 x2'
where the first and last terms are related to the variances of the two
individual equations. It would seem like they could be calculated (via stdp
for each equation) and added, and the result negated. Without working it
out in detail, I am not sure how - [- x2 V21 x1' - x1 V12 x2' ] relates to
what Alan seeks, but it seems like you ought to be able to get it out of
this.

Since Stata will calculate this stddp for any pair of equations, it would
be quite straightforward to have an ado do all the necessary predicts (for
both the stdp and stddp) and crank out the desired quantities, even for >2
eqns,
if they can be derived from the stddp magnitudes.

Kit
--------------------------------------------------------------------
Kit Baum, Faculty Micro Resource Center                  fmrc@bc.edu
Academic Technology Services, Boston College   http://www.bc.edu/ats
http://fmwww.bc.edu/FMRC/                http://fmwww.bc.edu/GStat/
*
*   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/
```