R : Pass argument to glm inside an R function

Question

I am trying to get used to scoping issues in R. I'd like to call the function glm() inside a function but it does not work, apparently for scoping reasons I did not manage to fix with the functions assign() or eval().

Here is a simplified version:

ao <- function (y, x, phi = seq (0,1,0.1), dataset, weights) {
    logLikvector <- rep(0,length(phi))  # vector of zeros to be replaced thereafter
    for (i in 1:length(phi)) {          # loop to use glm()   
        fit <- glm (y ~ x, data = dataset, family = binomial, weights = weights)         
        logLikvector[i] <- logLik(fit)      # get log likelihood
    }
    logLikvector
}

Now I want to use the function ao() on my dataset

    ao (y = Prop, x = Age, dataset = mydata, weights = Total) 

This does not work, but the following works:

ao (y = mydata$Prop, x = mydata$Age, dataset = mydata, weights = mydata$Total)

Does anyone know what to do ?

Any help would be greatly appreciated !!!

Btw, here is how to replicate my problem with the dataset I am using

library("MASS")
data(menarche)
mydata <- menarche
mydata$Prop <- mydata$Menarche / mydata$Total

Answer 1

Solution with substitute (@DWin suggestion).

function(y, x, dataset, weights){
  f <- substitute(glm(y~x, data=dataset, weights=weights, family=binomial))
  logLik(eval(f))
}

Answer 2

ao <- function (x, y, phi = seq (0,1,0.1), dataset, weights) {
    logLikvector <- rep(0,length(phi))
    x <- dataset[,substitute(x)]
    y <- dataset[,substitute(y)]
    weights <- dataset[,substitute(weights)]
        for (i in 1:length(phi)) {          # loop to use glm()
        fit <- glm (y ~ x, data = dataset, family = binomial, weights = weights)
        logLikvector[i] <- logLik(fit)      # get log likelihood
    }
    return(logLikvector)
}



library("MASS")
data(menarche)
mydata <- menarche
mydata$Prop <- mydata$Menarche / mydata$Total
ao(y = "Prop",x = "Age", dataset = mydata, weights = "Total")


[1] -55.37763 -55.37763 -55.37763 -55.37763 -55.37763 -55.37763
 [7] -55.37763 -55.37763 -55.37763 -55.37763 -55.37763

Answer 3

I suggest creating the formula with paste and calling the function with do.call.

ao <- function (y, x, phi = seq (0,1,0.1), dataset, weights) {
  logLikvector <- rep(0,length(phi))  # vector of zeros to be replaced thereafter
  for (i in 1:length(phi)) {          # loop to use glm()
    f <- as.formula(paste(y, x, sep="~"))
    fit <- do.call("glm", list(formula=f, data=as.name(dataset), 
                   family="binomial", weights=as.name(weights)))
    logLikvector[i] <- logLik(fit)      # get log likelihood
  }
  logLikvector
}

Then call it like this:

ao("Prop", "Age", dataset="mydata", weights="Total")

See https://stackoverflow.com/a/7668846/210673 for more details.