Bookmark and Share

Notice: On March 31, it was announced that Statalist is moving from an email list to a forum. The old list will shut down at the end of May, and its replacement, statalist.org is already up and running.


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

Re: st: Streg (gamma): discontinuous region encountered


From   Maarten Buis <maartenlbuis@gmail.com>
To   statalist@hsphsun2.harvard.edu
Subject   Re: st: Streg (gamma): discontinuous region encountered
Date   Fri, 18 Nov 2011 09:18:47 +0100

On Thu, Nov 17, 2011 at 10:51 PM, Jana von Stein wrote:
> I am trying to fit a Gamma model to determine which distribution to use in
> streg. The following happens:
<snip>
> Iteration 5:   log pseudolikelihood = -129.37384
> cannot compute an improvement -- discontinuous region encountered
>
> Does anyone know what might be going on? Or if there is some other way to do
> this?

The gamma model is a very flexible model, but that also means that it
requires data that contain a lot of information in order to estimate
such models. Flexible models are nice as they extract a lot of
information from the data, but it can only do so if the information is
there in the first place. More observations means more information,
but not all observations add the same amount of information. A dataset
with a given number of observations contains less information if it
has an categorical variable with one or more rare categories, or
similarly, a continuous variable with low variance. That dataset also
contains less information if some explanatory variables are more
highly correlated. Your result means that your data does not contain
enough information to estimate this model.

You might be lucky and there is just enough information is there, but
it has become such a difficult problem that you need to help Stata a
bit. For example try to scale the variables such that the values
aren't too big or too small, e.g. don't add annual income in
euros/dollars but annual income in 1000s of euros or dollars. Make
sure that for each variable the value zero is meaningful and
within/close to the range of the data, e.g. if your sample consists of
adults don't add year of birth but (year of birth - 1950). If you have
explanatory variables that are highly correlated with one another you
can look at -help orthog-.

Hope this helps,
Maarten

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


© Copyright 1996–2014 StataCorp LP   |   Terms of use   |   Privacy   |   Contact us   |   Site index