lme4 (version 0.999999-2)

refit: Re-fit a model to a new response vector

Description

This generic function fits a model to a new response vector. It typically is used as part of a simulation. For models fit by lmer or nlmer it is much faster to use this function to re-fit the model than to begin from the model formula.

Usage

refit(object, newresp, ...)

Arguments

object

a fitted model. Methods are defined for models fit by lmer and by nlmer

newresp

a new response vector - typically a numeric vector.

...

further arguments passed for some methods.

Value

• a fitted model similar to object

Details

For many model-fitting functions in R a large part of the execution time is taken up with converting a formula representation and the data argument into the numerical structures used to evaluate the parameter estimates. It is common in a simulation to use the same model specification and covariate data on many simulated responses. That is, the only thing that changes between model fits is the response vector. This generic function allows for the model specification to and covariate data to stay the same while the only the response vector is changed.

Examples

fm1 <- lmer(Reaction ~ Days + (Days|Subject), sleepstudy)
set.seed(54321)
system.time(simfe <-
        data.frame(t(apply(simulate(fm1, nsim = 100),
                           2, function(y) fixef(refit(fm1, y)))),
                   check.names = FALSE))
qqmath(~ `(Intercept)` + Days, simfe, outer = TRUE,
       ylab = NULL, scales = list(y = list(relation = "free")),
       layout = c(1,2), aspect = 1, type = c("g", "p"))
sapply(simfe, mean)
sapply(simfe, sd)