Re: st: oglm and heterogeneous choice models

[Richard Williams <richardwilliams.ndu@gmail.com>]

At 01:12 PM 10/24/2011, Rourke O'Brien wrote:
> I have a follow up question on heterogeneous choice models.
>
> I am interested in testing which variables significantly predict
> residual variance. I've tried many configurations of the "sw, pe(.05):
> oglm y x1 x2 x3 x4, eq2(x1 x2 x3 x4) flip" but have not been able to
> achieve convergence. Yet, when I use the gologit2 command and the
> autofit option, the model runs smoothly and I am told which covariates
> do not meet criteria for parallel line assumptions. I can then run the
> model using gologit2 and specify which predictors meet the parallel
> line assumptions and which do not. Is this an appropriate strategy? I
> am ultimately interested in testing for an interaction in a logistic
> model.

Is the dv dichotomous? These models can be difficult to estimate as 
is, and are even tougher with a dichotomous dv.

The oglm help recommends using the -lr- option of -sw-. The default 
-wald- option gets confused when the same variable is in both the 
variance and choice equations, i.e. it tests the variable in both 
equations when you only want it tested in the variance equation.

Either a brant test or gologit2 can identify variables that are 
problematic. You can try including those variables in the variance 
equation of a hetero model -- it may or may not work well.

Even though I programmed oglm to support sw, I am not crazy about its 
use. In my Stata Journal article (Stata Journal 10(4):540-567) I 
suggested you think of this as being like a diagnostic test. If the 
assumption of homogeneous errors seems to be violated, think about 
other ways to solve the problem, e.g. add a variable, add a squared 
term. I make the same advice for OLS models where hetero seems to be 
a problem -- see if there is some reasonable way to make the hetero 
go away by tweaking your model.


-------------------------------------------
Richard Williams, Notre Dame Dept of Sociology
OFFICE: (574)631-6668, (574)631-6463
HOME:   (574)289-5227
EMAIL:  Richard.A.Williams.5@ND.Edu
WWW:    http://www.nd.edu/~rwilliam

*
*   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/