[R] simulating future observations from heteroscedastic fits

Spencer Graves @pencer@gr@ve@ @end|ng |rom e||ect|vede|en@e@org
Thu Aug 31 08:27:06 CEST 2023

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)

(fit12 <- lme(y~1, data=DF3_2, random=~1|gp))
(fit22 <- lme(y~gp, data=DF3_2, random=~1|gp))
(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,
   with(pred, fit[1, 'fit'] +
          se.fit*rt(nSims, df))

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

	  Spencer Graves

More information about the R-help mailing list