R help - Sep 2006 - maximizing a likelihood function containing an integral

Hi, R Users;

I am trying to maximize a likelihood function which
contains an integral.  The integral contains the
unknown parameter as well.  I am trying to use the
following code to do the maximization:

ll<-function(b.vec){
       b0<-b.vec[1]
       b1<-b.vec[2]
       b2<-b.vec[3]
      p<-1/(1+exp(-b0-b1*z1-b2*x2))
      
lik1<-p^y*(1-p)^(1-y)*exp(-(z1^2+x2^2-2*rho*z1*x2)/(2*(1-rho^2)))
            log.lik1<-sum(log(lik1[1:n1]))
       log.lik2<-0
     for(j in (n1+1):n){
                    integrand<-function(u,B0,B1,B2){

exp(-y[j]*(B0+B1*u+B2*x2[j])-(u-rho*x2[j])^2/2)/(1+exp(B0+B1*u+B2*x2[j]))
}
  
log.lik2<-log.lik2+log(integrate(integrand,lower=1,upper=Inf,B0=b0,B1=b1,B2=b2)$integral)
   }
 
   log.lik<-log.lik1+log.lik2
}


 
start<-c(0,0,0)

nlminb(start,ll)

The error message is: 
Error in log(x) : Non-numeric argument to mathematical
function

Suggestions are welcome.

Thanks

Ming Ji

Sam Wong <qxsr <at> yahoo.com> writes:
> 
> 
> Hi, R Users;
> 
> I am trying to maximize a likelihood function which
> contains an integral.  The integral contains the
> unknown parameter as well.  I am trying to use the
> following code to do the maximization:
> 
  [snip]
> The error message is: 
> Error in log(x) : Non-numeric argument to mathematical
> function
> 

   The only obvious (?) mistake in the code is that
you refer to integrate(...)$integral -- the help page
for ?integrate says that $value is the estimate of
the integral, so you might be getting a NULL here.

However, the code you gave us is not reproducible/self-contained -- since I
don't
know what z1, z2, n1, rho are, I can't run it and see if that's the only
problem.

  cheers
    Ben Bolker