[R] simulating future observations from heteroscedastic fits

[Spencer Graves]

Hello, All:


	  I want to simulate future observations from fits to heteroscedastic 
data. A simple example is as follows:


(DF3_2 <- data.frame(y=c(1:3, 10*(1:3)),
                      gp=factor(rep(1:2, e=3))))


# I want to fit 4 models
# and simulate future observations from all 4:


fit11 <- lm(y~1, DF3_2)
fit21 <- lm(y~gp, DF3_2)

library(nlme)
(fit12 <- lme(y~1, data=DF3_2, random=~1|gp))
(fit22 <- lme(y~gp, data=DF3_2, random=~1|gp))
library(lme4)
(fit12r <- lmer(y~1+(1|gp), data=DF3_2, REML=FALSE))
(fit22r <- lmer(y~gp+(1|gp), data=DF3_2, REML=FALSE))

# I can simulate what I want for fit11 and fit21
# as follows:

simPred <- function(object, nSims=2){
   pred <- predict(object, DF3_2[6,], se.fit=TRUE,
                   interval='prediction')
   with(pred, fit[1, 'fit'] +
          se.fit*rt(nSims, df))
}
simPred(fit11)
simPred(fit21)


# How can I do the same with either fit12 and fit22
# or fit12r and fit22r?


	  Thanks,
	  Spencer Graves