[datatable-help] computation of variance covariance matrix for weighted least squares

[Dirkules]

Hi, 

I am trying to compute the variance covariance matrix in weighted least
squares. Using the lm function gives me a different result as when I am
computing it myself. Any idea what I am doing incorrectly? My calculation
for the beta coefficients seem correct though. Here is my code


library('foreign')

download.file('http://fmwww.bc.edu/ec-p/data/wooldridge/smoke.dta','smoke.dta',mode='wb')

smoke<-read.dta('smoke.dta')

######################################################################################
#computation of my weighting matrix "hhat"


lm.8.7<-lm(cigs ~ lincome + lcigpric + educ + age + agesq + restaurn,
data=smoke)

lres.u<-log(lm.8.7$residuals^2)

lm.8.7gls.u<-lm(lres.u ~ lincome + lcigpric + educ + age + agesq + restaurn,
data=smoke)

hhat<-exp(lm.8.7gls.u$fitted.values)

# estimating with weights hhat

lm.8.7gls<-lm(cigs ~ lincome + lcigpric + educ + age + agesq + restaurn,
              weights=1/hhat, data=smoke)

########################################################################################
  
# Doing the estimation myself

  u <- model.frame(cigs ~ lincome + lcigpric + educ + age + agesq +
restaurn, data=smoke)
  x <- model.matrix(u, smoke)
  y <- model.response(u)
  
  
  
  wg <- 1/hhat
  hh <- diag(wg)
  

  beta <- solve(t(x) %*% hh %*% x) %*% (t(x) %*% hh %*% y) # estimate of
beta
  
  dSigmaSq <- sum((y - x%*%beta)^2)/(nrow(x)-ncol(x))  # estimate of
sigma-squared
  
  
  vcm <- dSigmaSq  * solve(t(x) %*% hh %*% x) ## compute variance covariance
matrix
  
  t(beta)
  lm.8.7gls$coefficients
  vcm
  vcov(lm.8.7gls)



Many thanks!



--
Sent from: http://r.789695.n4.nabble.com/datatable-help-f2315188.html