Medflex within functions

[JonasBeal]

Dear medflex authors,

Thanks for this very convenient package. I have some problems when I try to use medflex functions within another function. This seems to me to be related to scoping problems, perhaps related to issues 6 or 14. However, the answers to these issues did not allow me to solve my problem satisfactorily. Here are examples.

1) Not working example

dta <- UPBdata
fmla <- UPB~att+negaff

return_direct_effect <- function(fmla, dta){
  output <- Reduce(paste, deparse(fmla[[2]]))#Left-hand variable as character
  input <- as.character(fmla[[3]][[2]])#Exposure variable as character

  expData <- neImpute(fmla, dta, family = gaussian)
  neMod <- neModel(as.formula(paste0(output, "~", input, "0+", input, "1")),
                   family = gaussian, expData = expData,
                   se = "bootstrap", nBoot = 10, progress=F)
  DE <- unname(neMod$neModelFit$coefficients[2])
  return(DE)
}
return_direct_effect(UPB~att+negaff, UPBdata)

Last command results in :

Error in eval(args$data, environment(stats::formula(fit))) : object 'dta' not found

Please note that naming my dta variable data as suggested in issue 14 for xFit seems to circumvent the problem but result in a new one

Error in data[[x]] : object of type 'closure' is not subsettable

2) Without function

If I remove the functional form it works fine:

dta <- UPBdata
fmla <- UPB~att+negaff
output <- Reduce(paste, deparse(fmla[[2]]))#Left-hand variable as character
input <- as.character(fmla[[3]][[2]])#Exposure variable as character

expData <- neImpute(fmla, dta, family = gaussian)
neMod <- neModel(as.formula(paste0(output, "~", input, "0+", input, "1")),
                   family = gaussian, expData = expData,
                   se = "bootstrap", nBoot = 10, progress=F)
DE <- unname(neMod$neModelFit$coefficients[2])

3) Force global variable

By the way, if I force my function/local variable to be global, putting the following code at the beginning of my function in 1), it works but it seems a little unsatisfactory

dta <<- dta
fmla <<- fmla

Thanks a lot for your insights and your time

[kaz-yos]

I seem to be having the same problem trying to run neWeight() within purrr:pmap() to examine different mediator model settings. I have a nested tibble with each row containing a formula, data tibble, and model fit. I wanted to run neWeight() in each row. The same data object not found error was encountered.

[kaz-yos]

It seems to stem from the discrepancy of the environment the model formula was defined and the environment in which the dataset resides. I'll try to resolve this.

eval(args$data, environment(stats::formula(fit)))