Creating categorical variables from mutually exclusive dummy variables

Question

My question regards an elaboration on a previously answered question about combining multiple dummy variables into a single categorical variable.

In the question p

Answer 1

Well, I think you can do it simply with ifelse, something like :

factor1 <- ifelse(is.na(conditionA), conditionB, conditionA)

Another way could be :

factor1 <- conditionA
factor1[is.na(factor1)] <- conditionB

And a third solution, certainly more pratical if you have more than two columns conditions :

factor1 <- apply(df[,c("conditionA","conditionB")], 1, sum, na.rm=TRUE)

Answer 2

I think this function gives you what you need (admittedly, this is a quick hack).

to_indicator <- function(x, grp)
{
    apply(tbl, 1,
          function (x)
          {
              idx <- which(!is.na(x))
              nm <- names(idx)
              if (nm %in% grp)
                x[idx]
              else
                NA
          })
}

And here is it's used with the example data you provide.

tbl <- read.table(header=TRUE, text="
conditionA    conditionB    conditionC     conditionD
NA            1             NA             NA
1             NA            NA             NA
NA            NA            1              NA
NA            NA            NA             1
NA            2             NA             NA
2             NA            NA             NA
NA            NA            2              NA
NA            NA            NA             2")
tbl <- data.frame(tbl)

(tbl <- cbind(tbl,
              factor1=to_indicator(tbl, c("conditionA", "conditionB")),
              factor2=to_indicator(tbl, c("conditionC", "conditionD"))))

Answer 3

Update (2019): Please use dplyr::coalesce(), it works pretty much the same.

My R package has a convenience function that allows to choose the first non-NA value for each element in a list of vectors:

#library(devtools)
#install_github('kimisc', 'muelleki')
library(kimisc)

df$factor1 <- with(df, coalesce.na(conditionA, conditionB))

(I'm not sure if this works if conditionA and conditionB are factors. Convert them to numerics before using as.numeric(as.character(...)) if necessary.)

Otherwise, you could give interaction a try, combined with recoding of the levels of the resulting factor -- but to me it looks like you're more interested in the first solution:

df$conditionAB <- with(df, interaction(coalesce.na(conditionA, 0), 
                                       coalesce.na(conditionB, 0)))
levels(df$conditionAB) <- c('A', 'B')