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

Re: AW: st: fractional logit convergence problem

From   Maarten buis <>
Subject   Re: AW: st: fractional logit convergence problem
Date   Wed, 18 Mar 2009 11:16:38 +0000 (GMT)

--- On Wed, 18/3/09, Jochen Späth <> wrote:
> Unfortunately, there is no perfect multicollinearity in my
> regressors: before doing the fractional logit I ran simple
> OLS on the same equations and none of the variables was
> dropped.

--- Maarten buis wrote:
> I would start with removing some of your covariates. Given 
> your describtion I would not be surprised if there is some
> perfect colinearity going on here. 

--- On Wed, 18/3/09, Jochen Späth wrote:
> > I try to fit a fractional logit model with 
> > -glm depvar indepvars, family(binomial) link(logit)
> > scale(x2)-, which uses Quasi-ML for the estimation.
> > My problem now is that the command does not converge.
> > What can I do? Would it help to try alternative
> > optimization> routines? Or change the distributional
> > assumption in -family()-?

I would still start with simplifying your model (i.e. remove covariates). That way you can at least identify which (combination of) variables is causing your problem. 

The binomial family and the logit link are usually very well behaved, so I don't think that is where the problem is. One exception would be perfect predictions, i.e. are there some combinations of covariates that produce all zeros or all ones. This causes problems with logistic regression, and I would expect it would also not play well with a fractional logit. 

-- Maarten

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


*   For searches and help try:

© Copyright 1996–2017 StataCorp LLC   |   Terms of use   |   Privacy   |   Contact us   |   What's new   |   Site index