It is true that for multinomial logit models, the marginal effects can change
sign depending on where the independent variables are being evaluated. I would
recommend "Regression Models for Categorical Dependent Variables Using Stata,
Revised Edition" by J. Scott Long, Jeremy Freese. They have provided a series
of useful commands for the post-estimation interpretation for discrete dependent
models. These can be located within Stata by typing -findit spostado- In
particular you may find -prchange- useful, which will compute the change in the
predicted probability for a discrete change in the independent variable.
Another possibility would be calculate the average of the marginal effects,
try -findit margin-
Hope this helps,
Scott
----- Original Message -----
From: "Tesfayi Gebre" <[email protected]>
To: <[email protected]>
Sent: Sunday, January 25, 2004 5:12 PM
Subject: st: Re: Interpretation of marginal effects in mlogit
> Dear Statalister,
>
> I used multinomial logit regression to estimate
> occupational choice model (I have 5 possibilities).
> However, the interpretation of both the estimated
> coefficients and the marginals is not straight
> forward. Econometrics books, such as Green (2000) warn
> against using the estimates for interpretation. For
> instance, Powers and Xie (2000) recommend using the
> odd-ratios for interpretation (since the marginals may
> not have the same sign as the coefficients).
>
> My question is how do you interpret the marginals from
> mlogit? If the results are difficult for
> interpretation, then what is the benefit of using
> mlogit?
>
> Best regards
> Tesfayi
>
>
*
* For searches and help try:
* http://www.stata.com/support/faqs/res/findit.html
* http://www.stata.com/support/statalist/faq
* http://www.ats.ucla.edu/stat/stata/
