Re: [S] Var-Covar Matrices of non-linear parameters

Prof Brian D Ripley (
Sun, 14 Jun 1998 08:54:46 +0100 (BST)

On 11 Jun 1998, Douglas Bates wrote:

> Pedro de Barros <> writes:
> > I am trying to find a way of obtaining the variance-covariance matrix for
> > the estimated parameters when running nlregb(), nls() or nlminb(),
> > equivalent to the "var" component of a glim() fit. (I want a simple way of
> > estimating confidence limits for the parameters).
> If you want to calculate confidence limits on the parameters in an nls
> fit you could use the approximate standard errors reported by the
> summary function. The object returned by the summary function applied
> to a fitted nls model has several components. One of these components
> is a matrix of parameter values, standard errors, and the t statistic.

[Examples deleted]

We think that the vcov function in our library MASS provides a
convenient way to do this. It has methods for nls and nlregb, that
for nls being equivalent to Douglas Bates' `manual' procedure.

For nlminb, the problem is not well-defined: it depends on what
you are minimizing. If it is minus a log-likelihood, vcov
has a suitable method.

Brian D. Ripley,        
Professor of Applied Statistics,
University of Oxford,             Tel:  +44 1865 272861 (self)
1 South Parks Road,                     +44 1865 272860 (secr)
Oxford OX1 3TG, UK                Fax:  +44 1865 272595

----------------------------------------------------------------------- This message was distributed by To unsubscribe send e-mail to with the BODY of the message: unsubscribe s-news