Bookmark and Share

Notice: On April 23, 2014, Statalist moved from an email list to a forum, based at

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

Re: st: MIXLOGIT: marginal effects

From   Arne Risa Hole <>
Subject   Re: st: MIXLOGIT: marginal effects
Date   Mon, 6 Feb 2012 14:03:39 +0000

Thanks Maarten. Just a small clarification: -mixlogit- allows for
multinomial outcomes while -xtmelogit- is for binary outcomes only so
the two are not substitutes in general. Another difference is that
-mixlogit- uses simulation to approximate the likelihood function,
while -xtmelogit- uses quadrature. For more information about
-mixlogit- see <>.

I disagree when it comes to marginal effects: I personally find them
much easier to interpret than odds-ratios. In the end the choice will
depend on your discipline and personal preference.


On 6 February 2012 13:29, Maarten Buis <> wrote:
> On Mon, Feb 6, 2012 at 1:57 PM, Davide Castellani wrote:
>> I am new to the mixlogit program. I was wondering how to work out marginal
>> effects ( and SEs of MEs).
> -mixlogit- is a user writen program, so per the Statalist FAQ (url at
> the bottom of every post on Statalist) you must say where you got it
> from. I will assume that you got it from the SSC archive, i.e. you
> used -ssc install mixlogit- to install -mixlogit-.
> It seems that -mixlogit- does not contain a standard predict function,
> so that means you cannot use any of the standard methods like
> -margins- or -mfx-.
> Since you did not say which version of Stata you are using, we will
> have to asume (again per the Statalist FAQ) that it is Stata 12. In
> that case you can just use the official -xtmelogit- command. This will
> allow you to use -margins- to get marginal effects at the average
> values (or other typical values) of the explanatory variables.
> However, many people seem to want average marginal effects, which is
> not the same. This is much harder to obtain.
> In general, I would just interpret the odds ratios. I do not
> understand why people go through a lot of trouble fitting a non-linear
> model and than undoing all that effort by only reporting marginal
> effects, which in effect reduces your non-linear model to a linear
> model. If you wanted a linear model, than why not fit a linear model?
> If you think a linear model is unacceptable in your case or in
> general, than so are marginal effects.
> Hope this helps,
> Maarten
> --------------------------
> Maarten L. Buis
> Institut fuer Soziologie
> Universitaet Tuebingen
> Wilhelmstrasse 36
> 72074 Tuebingen
> Germany
> --------------------------
> *
> *   For searches and help try:
> *
> *
> *

*   For searches and help try:

© Copyright 1996–2018 StataCorp LLC   |   Terms of use   |   Privacy   |   Contact us   |   Site index