[R] Problem with ordered probit model in MASS

[Prof Brian Ripley]

On Wed, 24 Jan 2007, justin bem wrote:

(Again: this is a duplicate post.)

> Dear all,
>
> I got this message, while using the polr function in MASS
>
>> EQ<-as.formula("dep~fpta+tcdv+cdta+cmta+prcd+patc+lactifs+excta")
>> Estim<-polr(EQ,don, subset=(cote!=0),method="probit",na.action=na.omit)
>
> Error in polr(EQ, don, subset = (cote != 0), method = "probit", na.action = na.omit) :
>        attempt for find suitable starting values failed
> In addition: Warning messages:
> 1: algorithm did not converge in: glm.fit(X, y1, wt, family = binomial("probit"), offset = offset)
> 2: fitted probabilities numerically 0 or 1 occurred in: glm.fit(X, y1, wt, family = binomial("probit"), offset = offset)
>
> how can I initialise starting values ?

What do you think the 'start' argument to polr() is for?

If you are asking how you find suitable values, I cannot help you as I 
know nothing about your problem, and failing to find starting values 
usually means that the model is very far from appropriate.

> Justin BEM
> Elève Ingénieur Statisticien Economiste
> BP 294 Yaoundé.
> Tél (00237)9597295.

-- 
Brian D. Ripley,                  ripley at stats.ox.ac.uk
Professor of Applied Statistics,  http://www.stats.ox.ac.uk/~ripley/
University of Oxford,             Tel:  +44 1865 272861 (self)
1 South Parks Road,                     +44 1865 272866 (PA)
Oxford OX1 3TG, UK                Fax:  +44 1865 272595