Re: st: Problem with the results of "mfx" command

[Maarten Buis <maartenlbuis@gmail.com>]

On Tue, Sep 4, 2012 at 12:40 PM, Kai Huang wrote:
> I have run an mlogit model on the probability of different types of employment and use the mfx command to estimate the marginal effect. One variable has its marginal effect being estimated, but the statistics of its parameter such as standard errors, z-values, confidence intervals are given as (.). The statistics in the original mlogit result are generated without any problems. Does anyone know what's wrong with this?

That can happen because a multinomial logit demands much more
information from the data than an ordered logit. If there isn't enough
information present in the data to support a multinomial logit model,
you can end up with such numerical problems.

Before giving up, I would first look at all the variable in your model
and make sure they all have a reasonable scale (e.g. annual income in
1000s of dollars not dollars) and that they are centered at meaningful
values within or near the range of the data. This can often help avoid
numerical problems.

If that does not work, I would look at small categories in your
dependent variable and see if you can merge that with an adjacent
category.

-- Maarten

---------------------------------
Maarten L. Buis
WZB
Reichpietschufer 50
10785 Berlin
Germany

http://www.maartenbuis.nl
---------------------------------

*
*   For searches and help try:
*   http://www.stata.com/help.cgi?search
*   http://www.stata.com/support/statalist/faq
*   http://www.ats.ucla.edu/stat/stata/