calculating standard errors using nlminb

["John Pitchard" <johnpitchard@googlemail.com>]

 

Dear all,

 

I'm trying to calculate individual standard errors for all the values within a  vector which has a restriction on it, namely they must add up to 1 - but not necessarily positive though, i.e

 

I've got a vector consisting of a, b, c, d

 

x <- c(a, b, c)

 

the aim is to the values of x that minimise some objective function using nlminb. Once I find these values I can calculate d as follows,

 

d <- 1-sum(x).

 

 

I want to obtain standard errors for a, b, c and d. Can this be achived from outputing the hessian matrix and then taking the square root of the diagonal elements? How would I calculate a standard error for component d as it is not part of the actual minimisation?

 

Many thanks in advance for any help.

 

Regards,

John