Question

I want this:

And I thought that passing na.action=na.pass to boxplot would let NA show up in
the grouping names. Here is some sample code:

#Build a fake dataset
set.seed(212012)
nn = 100
sample_data <- data.frame( score = c( rpois(nn, 1), rpois(nn, 2),
                                      rpois(nn, 1.5), rpois(nn, 3)),
                          category = c( rep(0, nn), rep(1, nn), 
                                        rep(2, nn), rep(NA, nn) ))   

boxplot( score ~ category, data=sample_data, na.action=na.pass )

But that produces this:

The ‘simple’ way to get what I want is the following code, but it’s not great for exploratory data analysis:

sample_data$category2 <- sample_data$category
sample_data$category2[ is.na(sample_data$category) ] <- 'NA'
boxplot( score ~ category2, data=sample_data )

Any hints from the R Guru’s out there? I was able to find out about na.pass from this more general discussion, and the origin of na.pass from Prof. Ripley here. But there doesn’t seem to be a distinction between missing data (NA’s) appearing in the data that will be split by the factor and missing data in the factor itself. Am I missing something simple, or is this more of a feature request?

Answer 1

boxplot( score ~ factor(category,exclude=NULL), data=sample_data)

the default factor behavior is exclude=NA. I assume internal to the boxplot call is a factor call if its not already a factor. This is just forcing the factorization to include your NA values.