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, statalist.org is already up and running.

# Re: st: nl: missing standar error

 From "JVerkuilen (Gmail)" To statalist@hsphsun2.harvard.edu Subject Re: st: nl: missing standar error Date Tue, 18 Sep 2012 08:36:42 -0400

```On Tue, Sep 18, 2012 at 4:15 AM, Maarten Buis <maartenlbuis@gmail.com> wrote:

> --- Jay wrote:
>>> You may have an empirically unidentified parameter.
>
>> Could you please give me a hint how to identify the problem? I
>
> In that case there is nothing you can do. To quote John Tukey (1986,
> p.74-75): "The combination of some data and an aching desire for an
> answer does not ensure that a reasonable answer can be extracted from
> a given body of data."

Yes, Tukey is right, but there are degrees of "nothing you can do."

First turn off robust standard errors! That might be your problem
right there. If you have reason to suppose that there's
heteroscedastity, bootstrapping is much safer.

Empirical unidentification happens when you have a model that is
formally identified but for which some combination of the data
provides no information. In an only slightly nonlinear model like
logistic regression you have perfect prediction on top of the usual
collinearity problems. Stata is kind enough to tell you about that.
True nonlinear models are plagued with identification problems that do
not occur in linear models that no programmer could anticipate. The
problem is thoroughly discussed in books on nonlinear regression, such
as Bates and Watts. You may have to:

-Suffer through the unidentification
-Shift models to something simpler.
-Impose assumptions to add information, e.g., a Bayesian prior, a
penalty term on the fit function (like a ridge), force a coefficient
estimate to

Chances are good it's doing damage elsewhere so I think suffering
though the unidentification is probably not a good idea until you've
determined why. Is the affected coefficient oddly large or very close
to 0?

Bates, DM, Watts, DG. (2007). Nonlinear Regression Analysis and Its
Applications. Wiley.

--
JVVerkuilen, PhD
jvverkuilen@gmail.com

"Out beyond ideas of wrong-doing and right-doing there is a field.
I'll meet you there. When the soul lies down in that grass the world
is too full to talk about." ---Rumi
*
*   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/
```