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]

Antwort: Re: Antwort: Re: st: Interaction terms

From   Justina Fischer <>
Subject   Antwort: Re: Antwort: Re: st: Interaction terms
Date   Mon, 4 Apr 2011 17:41:45 +0200


you are in principle, right (and we had this discussion before)

but: margins will give you the total effect for specified values of the interacting variable

in a non-linear model, this is quite neat as calculation by hand is possible but time-consuming schrieb: -----

Von: Richard Williams <>
Gesendet von:
Datum: 04.04.2011 05:18PM
Thema: Re: Antwort: Re: st: Interaction terms

At 04:08 AM 4/4/2011, Justina Fischer wrote:
>well, as suggested by Maarten, I would employ the mlogit estimator 
>and then use the margins command for interpretation of the interaction terms.

Of course, the margins command doesn't even report anything for the 
interaction effect (at least if you've correctly specified the model 
using factor variables.) To see what I mean, try something like

use "";, clear
logit  warmlt3 i.yr89 i.male i.white age ed prst i.yr89#c.age
margins, dydx(*)

That makes more sense to me anyway -- the interaction term can't 
change without the variables  that are used to compute it also 
changing. Basically, the old mfx command gave incorrect results for 
interaction effects, while the new margins command gives you no 
results. Still, people keep on asking for marginal effects and 
standard errors specific for the interaction term, and I don't quite 
understand why.

Richard Williams, Notre Dame Dept of Sociology
OFFICE: (574)631-6668, (574)631-6463
HOME:   (574)289-5227
EMAIL:  Richard.A.Williams.5@ND.Edu

*   For searches and help try:

*   For searches and help try:

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