Hi,
I want to bootstrap the likelihood ratio test for testing linear
restrictions on the parameters of a regression model. The aim is to perform
a size and power analysis. My question is, as indicated by the subject, how
do I approximate the null distribution? The code that I ran looks like
> boot.test <- function(data, indeces, maxit=20) {
+ y1 <- fit1+e1[indeces]
+ mod1 <- glm(y1 ~ X1-1, maxit=maxit)
+ y2 <- fit2+e2[indeces]
+ mod2 <- glm(y2~1, maxit=maxit)
+ t <- 2*(logLik(mod1)-logLik(mod2))
+ t
+ }> boot.lrtest <- boot(data=M1, statistic=boot.test, R=1999, maxit=100)
However, I suspect that I am not bootstrapping under the null hypothesis.
How do I perform the bootstrap under the null? I would really appriciate any
help, I feel rather lost.
/Anders
[[alternative HTML version deleted]]