找不到变量； 范围问题

Question

I'm running into some bizarre scoping (maybe?) problem... The MWE below calls RunSamples, prints the variable pr , then throws an error on the next line saying that pr does not exist. I can't seem to understand why the print function can find and print the variable pr , but lmer cannot. Any help would be appreciated. Also, If I change the pr argument to prot , the code runs fine.

require(lme4)
dat <-data.frame(value=1:10,Item=1:10,Protocol=rep(c("FFT","Data"),5))
RunSamples <- function(dat,form,pr) {
  rets <- list()
  print(pr)
  rets$Full <- lmer(update.formula(form,.~.),data=dat, subset= Protocol==pr )
  return( rets )
}
RunFullMain <- function(prot="CLASS") {
  ret <- list()

  form <-  value~0+Item
  ret$Item <- RunSamples(dat=dat,form=form    ,prot)
  return(ret)
}

Results <- list()
for (pp in c("FFT","CLASS","PLATO")) {
  Results[[pp]] <- RunFullMain(pp)
}

Answer 1

I'm not sure I can fix this, in the sense of having it Just Work, but I think I can explain what's happening and give you a reasonable workaround.

lmer does a variety of (probably overly) complicated things to try to evaluate the data and make sure that all the variables referred to in the formula are present. It tries to carry along all the variables in the data argument, and those present in the environment of the formula, but beyond that it has trouble. The proximal problem here is that your subset call refers to a variable that doesn't exist in either of those places. The easy workaround is to use the subset function instead of the subset argument:

RunSamples <- function(dat,form,pr) {
  rets <- list()
  rets$Full <- lmer(update.formula(form,.~.),
               data=subset(dat,Protocol==pr))
  return( rets )
}

so that the subsetting gets done right away, rather than waiting until the function is called from an environment where pr can't be found.

When I do this I then encounter a series of different problems, which are probably due to your setting up an unrealistically simple MWE ...

• Error: no random effects terms specified in the formula . (Changed 0+Item to 0+(1|Item) in the original formula)
• Error: number of levels of each grouping factor must be < number of observations (could probably deal with this one too ...)