Label Attributes for Atomic Vectors
Labels are a common construct in statistical software providing a human readable description of a variable. While variable names are succinct, quick to type, and follow a language's naming conventions, labels may be more illustrative and may use plain text and spaces.
R does not provide native support for labels. Some packages, however, have made this feature available. Most notably, the Hmisc package provides labelling methods for a number of different object. Due to design decisions, these methods are not all exported, and so are unavailable for use in package development.
The labelVector package supports labels for atomic vectors in a light-weight design that is suitable for use in other packages.
The package may be downloaded from CRAN using
install.packages("labelVector")
or from GitHub using
devtools::install_github("nutterb/labelVector")
When generating summaries for reports to be delivered to a non-technical audience, the variable names used in analytical code may not be adequately descriptive to the audience to provide the full context and meaning of the results. Variable labels are a compromise that may be inserted to clarify meaning to the audience without requiring excessively difficult variable names to be used in code.
In the table below, a linear model estimating gas mileage is given with terms taken from the variable labels.
library(labelVector)
mtcars <-
set_label(mtcars,
qsec = "Quarter mile time",
am = "Automatic / Manual",
wt = "Vehicle weight")
fit <- lm(mpg ~ qsec + am + wt,
data = mtcars)
# Create a summary table
res <- as.data.frame(coef(summary(fit)),
stringsAsFactors = FALSE)
res <- cbind(rownames(res), res)
rownames(res) <- NULL
names(res) <- c("term", "estimate", "se", "t", "p")
res$term <- as.character(res$term)
kable(res)| term | estimate | se | t | p |
|---|---|---|---|---|
| (Intercept) | 9.617781 | 6.9595930 | 1.381946 | 0.1779152 |
| qsec | 1.225886 | 0.2886696 | 4.246676 | 0.0002162 |
| am | 2.935837 | 1.4109045 | 2.080819 | 0.0467155 |
| wt | -3.916504 | 0.7112016 | -5.506882 | 0.0000070 |
In constrast, the following table replaces these term labels with longer, more human-readable terms that assist in the interpretation of the model.
res$term[-1] <- get_label(mtcars, vars = res$term[-1])
kable(res)| term | estimate | se | t | p |
|---|---|---|---|---|
| (Intercept) | 9.617781 | 6.9595930 | 1.381946 | 0.1779152 |
| Quarter mile time | 1.225886 | 0.2886696 | 4.246676 | 0.0002162 |
| Automatic / Manual | 2.935837 | 1.4109045 | 2.080819 | 0.0467155 |
| Vehicle weight | -3.916504 | 0.7112016 | -5.506882 | 0.0000070 |
Truthfully, I don't believe it to be likely that these labels will be retrieved and applied to summaries interactively, but they do become useful when writing functions to generate common summary tables.
Labels are easily set using the set_label function and retreived using the get_label function.
x <- 1:10
x <- set_label(x, "some integers")
x
get_label(x)Labels may also be assigned in a manner that is pipe-friendly.
mtcars2 <-
set_label(mtcars,
am = "Automatic",
mpg = "Miles per gallon",
cyl = "Cylinders",
qsec = "Quarter mile time")Whereas labelVector provides a similar functionality as is provided by the Hmisc package, and considering the widespread use of Hmisc, consideration is taken for the possibility that labelVector and Hmisc may need to work in the same environment. This is permissible since set_label and get_label both work on the label attribute of a vector and their names do not conflict with the label generic exported by Hmisc.