I'm pretty sure that this is a marginal effect, and the variance can be
determined via -mfx
Or see,
Graubard, B.I. and Korn, E.L. "Predictive margins for survey data"
Biometrics 55
about June, 1999
Bryan Sayer
Statistician, SSS Inc.
bsayer@s-3.com
-----Original Message-----
From: Anderson, Soren [mailto:SANDERSON@rff.org]
Sent: Thursday, February 13, 2003 5:22 PM
To: statalist@hsphsun2.harvard.edu
Subject: st: standard errors for dummy variables in logit
I am trying to calculate the standard error for the discrete change in
probability associated with a dummy variable in a logit model by hand. The
effect itself is given by the difference in predicted probability with and
without the dummy variable equal to 1:
E[P|d=1]-E[P|d=0],
where E[P] is the predicted probability and d is the dummy variable.
The variance of the difference should be given by
Var(E[P|d=1]) + Var(E[P|d=0]) - 2*Cov(E[P|d=1],E[P|d=0]).
Greene (2000, p.824) has a nice little formula for the variance of the
individual predicted probabilities (i.e., the first two terms). Does anybody
know the formula for the covariance of two predicted probabilities (i.e.,
the last term)?
Soren Anderson
Resources for the Future
1616 P Street NW
Washington, DC 20036
202.328.5105
sanderson@rff.org
*
* 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/