Psych's reverse.code function producing NAs in R

Question

When using reverse.code in R, the values in my ID column (which are not meant to be reversed) turn into NA once the ID value exceeds 999 (I have 10,110 observations).

Does anyone know if there is anything I can do to fix this?

Is there another function I can use to reverse these items without loosing data?

Here is my code:

library(psych)
keys <- c(1,-1,-1,-1) #Where column 1 = ID and the rest are my variables to be reversed
rev_dat2 <- reverse.code(keys, rev_dat)

Thanks!

Answer 1

Here is the relevant line of the source code of reverse.code(), where new is the object holding the reverse-coded data:

new[abs(new) > 999] <- NA

As you can see, setting values larger than 9999 to missing is hard-coded into the routine. You could write a new version of the function that didn't do that. For example, in the function below, we just make a much larger threshold:

my.reverse.code <- function (keys, items, mini = NULL, maxi = NULL) 
{
  if (is.vector(items)) {
    nvar <- 1
  }
  else {
    nvar <- dim(items)[2]
  }
  items <- as.matrix(items)
  if (is.null(maxi)) {
    colMax <- apply(items, 2, max, na.rm = TRUE)
  }
  else {
    colMax <- maxi
  }
  if (is.null(mini)) {
    colMin <- apply(items, 2, min, na.rm = TRUE)
  }
  else {
    colMin <- mini
  }
  colAdj <- colMax + colMin
  if (length(keys) < nvar) {
    temp <- keys
    if (is.character(temp)) 
      temp <- match(temp, colnames(items))
    keys <- rep(1, nvar)
    keys[temp] <- -1
  }
  if (is.list(keys) | is.character(keys)) {
    keys <- make.keys(items, keys)
    keys <- diag(keys)
  }
  keys.d <- diag(keys, nvar, nvar)
  items[is.na(items)] <- -99999999999
  reversed <- items %*% keys.d
  adj <- abs(keys * colAdj)
  adj[keys > 0] <- 0
  new <- t(adj + t(reversed))
  new[abs(new) > 99999999999] <- NA
  colnames(new) <- colnames(items)
  colnames(new)[keys < 0] <- paste(colnames(new)[keys < 0], 
                                   "-", sep = "")
  return(new)
}

The reason they used a numeric value threshold is that for the recoding they do to work, they needed all values to be numeric. So, they set missing values to -999 and then later turn them back into missing values. The same is done above, but with a lot bigger number.

keys <- c(1,-1,-1,-1) #Where column 1 = ID and the rest are my variables to be reversed
rev_dat <- data.frame(
  id = 9998:10002, 
  x = 1:5, 
  y = 5:1, 
  z = 1:5
)
library(psych)
reverse.code(keys, rev_dat)
#      id x- y- z-
# [1,] NA  5  1  5
# [2,] NA  4  2  4
# [3,] NA  3  3  3
# [4,] NA  2  4  2
# [5,] NA  1  5  1
my.reverse.code(keys, rev_dat)
#         id x- y- z-
# [1,]  9998  5  1  5
# [2,]  9999  4  2  4
# [3,] 10000  3  3  3
# [4,] 10001  2  4  2
# [5,] 10002  1  5  1