Bookmark and Share

Notice: On March 31, it was announced that Statalist is moving from an email list to a forum. The old list will shut down on April 23, and its replacement, is already up and running.

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

Re: st: Interpreting marginal effects for binary variables in multinomial logit

From   David Hoaglin <>
Subject   Re: st: Interpreting marginal effects for binary variables in multinomial logit
Date   Thu, 14 Jun 2012 07:52:51 -0400


Whether ceteris paribus can apply to your model depends on how the
data cover the various combinations of the binary independent
variables.  If you want to examine the effect of changing one of those
independent variables while holding the other independent variables
constant, you need to do it at combinations of those variables where
you have data for both values of the variable that you're changing.
It comes down to examining the predicted values across the whole set
of combinations of the independent variables (in your data).

For various types of regression models, the common interpretation
(found, unfortunately, in many textbooks) that holds the other
independent variables constant is generally an oversimplification.
What regression actually does is adjust for the contributions of the
other independent variables.  It may not be possible to hold them all
constant or, depending on the data, to hold them constant in any
meaningful place (as Austin pointed out).

David Hoaglin

On Thu, Jun 14, 2012 at 7:19 AM, Julian Runge <> wrote:
> Thanks for your comments. I agree on the "awkwardness" of fixing
> binary covariates at the mean.
*   For searches and help try:

© Copyright 1996–2016 StataCorp LP   |   Terms of use   |   Privacy   |   Contact us   |   Site index