Dear Professor, Thank you very much for your quick and accurate answer,
and sorry for asking you about so simple stuff: getting back to basics!
Now I finally get it: the estimation of marginal effects in non-linear
models is quite controversial: before people used to estimate marginal
effects at the mean, now they tend to prefer average marginal
effects...; and they are indeed approximations.
And I had totally forgot the "rules of thumb" to compare probit, logit
and linear probability model's coefficients.
Best Regards,
Miguel García-Posada Department of Economics
Universidad Carlos III de Madrid (Spain)
Gary King <king(a)harvard.edu> dijo:
Miguel, marginal effects are usually defined as the
best linear
approximation to a nonlinear functional form. To do that for logit, or
relogit, just divide the coefficients by 4. But if you want to know the
exact change in the probability of Y=1 for a change from X1 being one value
to X1 being another value, holding constant the other X's at some given
values, then you want a first difference. There's no reason to use the
approximation here in marginal effects. Logit is a nonlinear functional
form; we don't need a linear approximation to it. the first difference is
of immediate substantive importance, interpretable by anyone without
statistics experience. seems more valuable to me.
see also King, Gary, Michael Tomz, and Jason Wittenberg. "Making the Most of
Statistical Analyses: Improving Interpretation and Presentation." American
Journal of Political Science 44 (2000): 341-355.
http://j.mp/jYibQd
Gary
--
*Gary King* - Albert J. Weatherhead III University Professor - Director,
IQSS - Harvard University
GKing.Harvard.edu <http://gking.harvard.edu/> - King(a)Harvard.edu -
@kinggary<http://twitter.com/kinggary>- 617-500-7570 - Asst 495-9271 -
Fax 812-8581
On Wed, Jun 8, 2011 at 12:19 PM, GARCIA-POSADA GOMEZ, MIGUEL <
100026381(a)alumnos.uc3m.es> wrote:
Dear Professor King,
I am a Phd student of Economics at Universidad Carlos III, Spain. I was
wondering if you could tell me:
1.Why you (and your coauthors) decided not to include marginal effects as a
quantity of interest in the software of Relogit, nor mentioning them in your
related papers. As you know, it?s probably the most used quantity of
interest in Economics; in fact, I?ve found a number of papers that, in order
to circumvent that ?problem? of Relogit, they first estimate the regressions
via both an ordinary logit and a relogit, compare the coefficients of the
two regressions and argue that, since they don?t differ ?substantially?,
they stick to the ordinary logit because it allows the computation of
marginal effects.
2.To the question of another student ?How to get the marginal effects after
running Relogit?? you answered ?better than marginal effects are first
differences, which are exact.? Although I have thoroughly read the 2 papers
you wrote with L.Zeng in 2001 about Relogit, I don?t mananage to understand
your answer.
I would greatly appreciate your help regarding these issues.
Best regards,
Miguel García-Posada Department of Economics
Universidad Carlos III de Madrid (Spain)
--
relogit mailing list served by Harvard-MIT Data Center
List Address: relogit(a)lists.gking.harvard.edu
Subscribe/Unsubscribe:
http://lists.gking.harvard.edu/?info=relogit
--
GARCIA-POSADA GOMEZ, MIGUEL
WebCartero
Universidad Carlos III de Madrid
--
relogit mailing list served by Harvard-MIT Data Center
List Address: relogit(a)lists.gking.harvard.edu
Subscribe/Unsubscribe: