On Aug 31, 2004, at 2:33 AM, Jose wrote:
Thanks very much. The problem is that stdp generates the standard
error for
the fitted values of X. This is also commonly referred to as the
standard
error of the observation's covariate pattern xjb. In the panel data I'm
using, this error is unique to each observation fitted. The reason I
use mfx
compute, however, is because I want to isolate the effect of a discrete
change in one variable on the dependent variable. If I x changes from
4 to
5, and y increases by 20, how do I get the standard error for this
particular change?. Unfortunately, I don't know how to get stata to get
stata to do this for E[y | x].
I don't get it. If you are using standard linear regression, you get a
point and interval estimate on the coefficient of X. It doesn't matter
what value X takes on (4 or 400) since the point and interval estimates
do not depend on the value of X. So \partial Y / \partial X is
\hat{\beta}, and \hat{beta} has a confidence interval, since it is a
random variable. There is thus a confidence interval for the random
variable \hat{Y}, given by multiplying dX by the upper and lower limits
of the confidence interval for \hat{\beta}. That gives you a range of
values on the Y scale that would arise by changing that particular X by
dX.
mfx compute, applied after regress, gives the CI for each \hat{\beta},
but in terms of Z, not t. It does not tell you anything that regress
does not (apart from the mean values of the regressors). But what are
you trying to compute from the regression?
Kit
*
* 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/
