Problems with boot and optim

 
I am trying to bootstrap the parameters for a model that is estimated
through the optim() function and find that when I make the call to boot,
it runs but returns the exact same estimate for all of the bootstrap
estimates.  I managed to replicate the same problem using a glm() model
but was able to fix it when I made a call to the variables as data frame
by their exact names.  But no matter how I refer to the variables in the
het.fit function (see below) I get the same result.  I could bootstrap
it with the sample command and a loop, but then the analysis in the next
step isn't as nice
 
The code for the likelihood and the call to boot is below.  I have tried
numerous other permutations as well.
 
I am using R 1.9.1 on Windows XP pro.
 
Thanks
 
Luke Keele
 
 
#Define Likelihood
lik.hetprobit <-function(par, X, Y, Z){
 
#Pull Out Parameters
Y <- as.matrix(y)
X <- as.matrix(x)
Z <- as.matrix(z)
K <-ncol(X)
M <-ncol(Z)
b <- as.matrix(par[1:K])
gamma <- as.matrix(par[K+1:M])
 
mu <- (X%*%b)
sd <- exp(Z%*%gamma)
 
mu.sd <-(mu/sd)
 
#Form Likelihood
 
   log.phi <- pnorm(ifelse(Y == 0, -1, 1) * mu.sd, log.p = TRUE)
   2 * sum(log.phi)
}
 
y <- as.matrix(abhlth)
x <- as.matrix(reliten) 
ones = rep(1, nrow(x)) 
x = cbind(ones,x) 
z = as.matrix(abinfo)
 
data.het <- as.matrix(cbind(y,x,z))
 
het.fit <- function(data){
    mod <- optim(c(1,0,0), lik.hetprobit, Y=data$y, X=data$x, Z=data$z,
method="BFGS", 
                   control=list(fnscale=-1), hessian=T)
    c(mod$par)
    }
case.fun <- function(d,i)
het.fit(d[i,])
 
het.case <- boot(data.het, case.fun, R=50)
Luke Keele
Post-Doctoral Fellow in Quantitative Methods
Nuffield College, Oxford University
Oxford, UK
 

	[[alternative HTML version deleted]]