order {base}R Documentation

Ordering Permutation


order returns a permutation which rearranges its first argument into ascending or descending order, breaking ties by further arguments. sort.list is the same, using only one argument.
See the examples for how to use these functions to sort data frames, etc.


order(..., na.last = TRUE, decreasing = FALSE)

sort.list(x, partial = NULL, na.last = TRUE, decreasing = FALSE,
          method = c("shell", "quick", "radix"))



a sequence of numeric, complex, character or logical vectors, all of the same length, or a classed R object.


an atomic vector.


vector of indices for partial sorting. (Non-NULL values are not implemented.)


logical. Should the sort order be increasing or decreasing?


for controlling the treatment of NAs. If TRUE, missing values in the data are put last; if FALSE, they are put first; if NA, they are removed (see ‘Note’.)


the method to be used: partial matches are allowed. The default is "shell" except for some special cases: see ‘Details’. For details of methods "shell" and "quick", see the help for sort.


In the case of ties in the first vector, values in the second are used to break the ties. If the values are still tied, values in the later arguments are used to break the tie (see the first example). The sort used is stable (except for method = "quick"), so any unresolved ties will be left in their original ordering.

Complex values are sorted first by the real part, then the imaginary part.

The sort order for character vectors will depend on the collating sequence of the locale in use: see Comparison.

The default method for sort.list is a good compromise. Method "quick" is only supported for numeric x with na.last = NA, and is not stable, but will be substantially faster for long vectors. Method "radix" is only implemented for integer x with a range of less than 100,000. For such x it is very fast (and stable), and hence is ideal for sorting factors—as from R 3.0.0 it is the default method for factors with less than 100,000 levels. (This is also known as counting sorting.)

partial = NULL is supported for compatibility with other implementations of S, but no other values are accepted and ordering is always complete.

For a classed R object, the sort order is taken from xtfrm: as its help page notes, this can be slow unless a suitable method has been defined or is.numeric(x) is true. For factors, this sorts on the internal codes, which is particularly appropriate for ordered factors.


An integer vector unless any of the inputs has 2^31 or more elements, when it is a double vector.


In programmatic use it is unsafe to name the ... arguments, as the names could match current or future control arguments such as decreasing. A sometimes-encountered unsafe practice is to call do.call('order', df_obj) where df_obj might be a data frame: copy df_obj and remove any names.


sort.list can get called by mistake as a method for sort with a list argument: it gives a suitable error message for list x.

There is a historical difference in behaviour for na.last = NA: sort.list removes the NAs and then computes the order amongst the remaining elements: order computes the order amongst the non-NA elements of the original vector. Thus

   x[order(x, na.last = NA)]
   zz <- x[!is.na(x)]; zz[sort.list(x, na.last = NA)]

both sort the non-NA values of x.

Prior to R 3.1.0 method = "radix" was only supported for non-negative integers.


Becker, R. A., Chambers, J. M. and Wilks, A. R. (1988) The New S Language. Wadsworth & Brooks/Cole.

Knuth, D. E. (1998) The Art of Computer Programming, Volume 3: Sorting and Searching. 2nd ed. Addison-Wesley.

See Also

sort, rank, xtfrm.



(ii <- order(x <- c(1,1,3:1,1:4,3), y <- c(9,9:1), z <- c(2,1:9)))
## 6  5  2  1  7  4 10  8  3  9
rbind(x, y, z)[,ii] # shows the reordering (ties via 2nd & 3rd arg)

## Suppose we wanted descending order on y.
## A simple solution for numeric 'y' is
rbind(x, y, z)[, order(x, -y, z)]
## More generally we can make use of xtfrm
cy <- as.character(y)
rbind(x, y, z)[, order(x, -xtfrm(cy), z)]

## Sorting data frames:
dd <- transform(data.frame(x, y, z),
                z = factor(z, labels = LETTERS[9:1]))
## Either as above {for factor 'z' : using internal coding}:
dd[ order(x, -y, z), ]
## or along 1st column, ties along 2nd, ... *arbitrary* no.{columns}:
dd[ do.call(order, dd), ]

set.seed(1)  # reproducible example:
d4 <- data.frame(x = round(   rnorm(100)), y = round(10*runif(100)),
                 z = round( 8*rnorm(100)), u = round(50*runif(100)))
(d4s <- d4[ do.call(order, d4), ])
(i <- which(diff(d4s[, 3]) == 0))
#   in 2 places, needed 3 cols to break ties:
d4s[ rbind(i, i+1), ]

## rearrange matched vectors so that the first is in ascending order
x <- c(5:1, 6:8, 12:9)
y <- (x - 5)^2
o <- order(x)
rbind(x[o], y[o])

## tests of na.last
a <- c(4, 3, 2, NA, 1)
b <- c(4, NA, 2, 7, 1)
z <- cbind(a, b)
(o <- order(a, b)); z[o, ]
(o <- order(a, b, na.last = FALSE)); z[o, ]
(o <- order(a, b, na.last = NA)); z[o, ]

##  speed examples for long vectors:
x <- factor(sample(letters, 1e6, replace = TRUE))
system.time(o <- sort.list(x)) ## 0.4 secs
system.time(o <- sort.list(x, method = "quick", na.last = NA)) # 0.1 sec
system.time(o <- sort.list(x, method = "radix")) # 0.01 sec
xx <- sample(1:26, 1e7, replace = TRUE)
system.time(o <- sort.list(xx, method = "radix")) # 0.1 sec
xx <- sample(1:100000, 1e7, replace = TRUE)
system.time(o <- sort.list(xx, method = "radix")) # 0.5 sec
system.time(o <- sort.list(xx, method = "quick", na.last = NA)) # 1.3 sec

[Package base version 3.3.0 Index]