Statalist


[Date Prev][Date Next][Thread Prev][Thread Next][Date Index][Thread Index]

st: Problems with average marginal effects (margeff)


From   Fardad Zand <fardad.zand@gmail.com>
To   statalist@hsphsun2.harvard.edu
Subject   st: Problems with average marginal effects (margeff)
Date   Sat, 5 Sep 2009 13:27:59 +0200

Dear Listers,

I'm running clogit; I couldn't find any available command to compute
"average marginal effect" after clogit (I checked mfx, mfx2, margeff).

Then I replaced clogit with DV logit (including dummy variables).

Thus, I could run margeff.

When one types -findit margeff in Stata, at least two different
versions are found:

1- st0086 from http://www.stata-journal.com/software/sj5-3: Stata Journal
After installation, this version produces results with comparable
Z-values and significance as the -logit

2- margeff from http://web.uni-corvinus.hu/bartus/stata: Bartus site
After installation, this version produces results comparable to the
previous installation (in terms of coefficients) but different
(usually improved) Z-values and thus significances. This time the
output is such that some of the variables which yielded insignificant
results when running -logit, gain significance.

There questions:

1- Is anyone aware of a post-estimation command to compute "average
marginal effect" after clogit (with predict (pu0), i.e. d
pr(y=1)/dx))?
2- If not, which of the above two installations is the most accurate
and reliable one to report?
3- In general, is it theoretically possible to obtain an insignificant
odds ratio but a significant marginal effect for a (dummy) explanatory
variable?

Thank you so much for your kind support.

Bests,
Fardad
*
*   For searches and help try:
*   http://www.stata.com/help.cgi?search
*   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