Question

I am trying to create a loop that will perform a series of analyses. I am using geeglm from geepack, which fails if there are any null values. Creating a subset solves this, but do not seem to be able to set the subset dynamically based on a changing variable.

while (j <= y.num) {

     strSubset = as.character(df.IV$IV[j])  #Gives column name in quotes
     df.data.sub = subset(df.data, strSubset>=0)

 #subset dataset is not created

 # analyses on subset take place

    j = j + 1
 }

If I type the variable name in the formula it works, so I assume that I am not creating the variable in a manner that allows it to be evaluated in the subset function. Any help would be greatly appreciated!

Reproducible example:

# dataset
age<-18:29
height<-58:69
df.ex=data.frame(age=age,height=height)
df.ex[4,1]<-NA

# dataset of columns that will be used for analysis
values<-c("age", "height")
df.variables=data.frame(values)

 # Age column has a null (NA) value.  The row must be removed for the analysis to run
 # explicit creation  
df.ex.sub.explicit<-subset(df.ex, age >= 0)
dim(df.ex.sub.explicit) #11 obs of 2 variables


i=1
strFilter=as.character(df.variables$values[i])
df.ex.sub.passvar<-subset(df.ex,strFilter>=0)
dim(df.ex.sub.explicit) #12 obs of 2 variables

Answer 1

I would suggest:

df.ex=data.frame(age=18:29,height=58:69)
df.ex[4,1]<-NA

It’s a little easier to store this list of variables as a character vector, unless you need the variables to be coupled with other information about the variables …

df.variables <- c("age", "height")

for (i in seq_along(df.variables)) {
  vname <- df.variables[i]  ## get variable name
  df.ex.sub.passvar <- df.ex[!is.na(df.ex[[vname]]),]
  print(dim(df.ex.sub.passvar))
}

subset and $ are great for interactive use, but for programming it is probably best to use “machine-style” indexing with [ and [[ … also, you need to use is.na() to test for NA values. subset() has a quirk in that it will drop values for which the result of the test is either FALSE or NA, but it is probably clearer to use the explicit test.