[R] polr() error message wrt optim() and vmmin

[Prof Brian Ripley]

On Sat, 10 Nov 2007, Anders Schwartz Corr wrote:

> I'm getting an error message using polr():
>
> Error in optim(start, fmin, gmin, method = "BFGS", hessian = Hess, ...) :
>         initial value in 'vmmin' is not finite
>
> The outcome variable is ordinal and factored, and the independant variable
> is continuous. I've checked the source code for both polr() and optim()
> and can't find any variable called "vmmin" as referenced in the error
> message.

vmmin is the C code for optim(method="BFGS"), so I don't know how you 
could have missed it.  See line 304 of src/main/optim.c!

What you did not say is that you were using polr(method="cauchit"), where 
finding starting values is notoriously difficult.  Either supply starting 
values or call polr(..., Hess=TRUE) if you know you are going to be 
calling summary() on the result.


>
> Any suggestions welcome.
>
> Thanks,
>
> Anders
>
>
>> summary(data$ordaodb)
>    0    1    2    3    4
> 8228 3505 1755  559 1757
>> summary(data$gpconc)
>    Min. 1st Qu.  Median    Mean 3rd Qu.    Max.
>   0.138   0.280   0.313   0.305   0.332   0.515
>> formula<-data$ordaodb~data$gpconc
>> propensityordaodb.plr <- polr(formula, method="cauchit")
>> propensityordaodb.plr
> Call:
> polr(formula = formula, method = "cauchit")
>
> Coefficients:
> data$gpconc
>       8.508
>
> Intercepts:
>   0|1   1|2   2|3   3|4
> 2.651 3.569 4.593 5.304
>
> Residual Deviance: 40009.85
> AIC: 40019.85
>> summary(propensityordaodb.plr)
>
> Re-fitting to get Hessian
>
> Error in optim(start, fmin, gmin, method = "BFGS", hessian = Hess, ...) :
>         initial value in 'vmmin' is not finite

-- 
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