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

RE: st: dprobit and lincom

From   Maarten buis <[email protected]>
To   [email protected]
Subject   RE: st: dprobit and lincom
Date   Fri, 22 Jan 2010 02:53:39 -0800 (PST)

--- On Thu, 21/1/10, Melanee Thomas wrote:
> what's the appropriate syntax to use to get an
> interpretable value from lincom after using dprobit using
> mfx? I'm assuming such syntax exists and that others have
> used it successfully..... 

It sounds like you want to interpret an interaction effect in a probit
model, and you want to interpret the effects as marginal effects. That 
is actually a surprisingly complicated problem, see for example the
article by Edward Norton and collegues (2004). A simple alternative is
to use -logit- rather than -probit- and interpret the odds ratios, as
is discussed here: <>.
If you want to stick to marginal effects and -probit- you can compute
those as in the example below, or you can take a look the -inteff-
command by Edward Norton and collegues (see the reference at the 
bottom of this post, and -findit inteff-).

*----------- begin example -----------------------
sysuse auto, clear
gen bad = rep78 <= 3 if rep78 < .
gen mpgXbad = mpg*bad
probit foreign mpg weight bad mpgXbad

sum mpg if e(sample), meanonly
local xb = _b[_cons] + _b[mpg]*r(mean)
sum weight if e(sample), meanonly
local xb = `xb'+_b[weight]*r(mean)
nlcom (good: normalden(`xb')*_b[mpg]) ///
      (bad: normalden(`xb' + _b[bad])*(_b[mpg]+_b[mpgXbad]))
*-------------- end example --------------------

--- Austin Nichols wrote:
> You should really not use -mfx- or -dprobit- at all, as the marginal
> effect at the mean is not informative for most purposes

I think that is a bit strong. When it comes to interpreting these 
interaction effects with marginal effects I would probably use both, 
as the average marginal effects not only include differences in 
effect but also differences in the distribution of the controll 
variables. The marginal effects at the means, by necesity control
for this. So by looking at both I get an idea of how much is due 
differences in effects and how much is due to differences in 
distributions of the controll variables. 

Hope this helps,

Edward C. Norton, Hua Wang, and Chunrong Ai (2004) Computing interaction effects and standard errors in logit and probit models. The Stata Journal, 4(2): 154--167.

Maarten L. Buis
Institut fuer Soziologie
Universitaet Tuebingen
Wilhelmstrasse 36
72074 Tuebingen


*   For searches and help try:

© Copyright 1996–2024 StataCorp LLC   |   Terms of use   |   Privacy   |   Contact us   |   What's new   |   Site index