Calling mlogit() from inside another function, scoping problem with variables when using attach

I need to call the mlogit() R function from inside another function.

This is a function for demonstrative purposes:

#-------------------------
# DEMO FUNCTION
#-------------------------
# f = formula (string)
# fData = data.frame
# cVar = choice variable (string)
# optVar = alternative variable (string)
##########################
mlogitSum <- function(f, fData,  cVar="choice", optVar="option"){
  library(mlogit)
  r2 <- mlogit(as.formula(f), shape = "long", data = fData, alt.var=optVar, choice = cVar)
  return(summary(r2))
}

Apparently there is an environment problem, so that variables not declared globally are not found by the mlogit() function as arguments.

This example doesn't work:

mydata <- read.csv(url("http://www.ats.ucla.edu/stat/r/dae/mlogit.csv"))
attach(mydata)
library(mlogit)
mydata$brand<-as.factor(mydata$brand)
mlData<-mlogit.data(mydata, varying=NULL, choice="brand", shape="wide")

myFormula <-"brand~1|female+age"
var1 <- "brand"
var2 <- "alt"
mlogitSum(myFormula, fData = mlData, var1, var2)

While if the variables are assigned in the main environment it works:

mydata <- read.csv(url("http://www.ats.ucla.edu/stat/r/dae/mlogit.csv"))
attach(mydata)
library(mlogit)
mydata$brand<-as.factor(mydata$brand)
fData<-mlogit.data(mydata, varying=NULL, choice="brand", shape="wide")

myFormula <-"brand~1|female+age"
cVar <- "brand"
optVar <- "alt"
mlogitSum(myFormula, fData, cVar, optVar)

Alternatively it works if I assign the variables globally from inside the function

#-------------------------
# DEMO FUNCTION
#-------------------------
# f = formula (string)
# fData = data.frame
# cVar = choice variable (string)
# optVar = alternative variable (string)
##########################
mlogitSum_rev <- function(f, fData,  cVar="choice", optVar="option"){
  fData<<-fData
  cVar<<-cVar
  optVar<<-optVar
  #return(head(lcmData))
  library(mlogit)
  #mi serve per poi estrarre model.matrix(r2), per il resto sarebbe ridondante
  r2 <- mlogit(as.formula(f), shape = "long", data = fData, alt.var=optVar, choice = cVar)
  return(summary(r2))
}

mydata <- read.csv(url("http://www.ats.ucla.edu/stat/r/dae/mlogit.csv"))
attach(mydata)
library(mlogit)
mydata$brand<-as.factor(mydata$brand)
mlData<-mlogit.data(mydata, varying=NULL, choice="brand", shape="wide")

myFormula <-"brand~1|female+age"
var1 <- "brand"
var2 <- "alt"
mlogitSum_rev(myFormula, mlData, var1, var2)

Any idea on how to avoid to assign the variables globally?

Solution

tl;dr this appears to be a bug in mlogit, which you can fix yourself (see below) or ask the maintainer to fix.

Deep inside mlogit, the function tries to evaluate the data as follows:

nframe <- length(sys.calls()) ## line 11
... 
data <- eval(mldata, sys.frame(which = nframe))  ## line 44

This is moderately sophisticated messing about with R's scoping structures -- it's trying to evaluate mldata in the frame one above the current frame, and it will fail if someone does something tricky (but perfectly reasonable!) like call mlogit from within a function.

I solved the problem (sort of!) by running fix(mlogit), which will dump you into an editor and allow you to modify the function. I changed line 44 to

data <- eval(mldata, parent.frame())

after which the code seemed to work.

If this works for you, you can either (1) fix() mlogit every time you need to use it: (2) download a copy of the source (.tar.gz) package, modify it, and install it; or (3) [preferably!] contact the package maintainer, let them know about the issue, and ask them to release a patched version ...

PS depending on your general data analysis protocol, you may want to get out of the habit of using attach: Why is it not advisable to use attach() in R, and what should I use instead?