base (version 3.4.1)

Extract: Extract or Replace Parts of an Object


Operators acting on vectors, matrices, arrays and lists to extract or replace parts.


x[i, j, … , drop = TRUE]
x[[i, exact = TRUE]]
x[[i, j, …, exact = TRUE]]
getElement(object, name)

x[i] <- value x[i, j, …] <- value x[[i]] <- value x$name <- value

Atomic vectors

The usual form of indexing is [. [[ can be used to select a single element dropping names, whereas [ keeps them, e.g., in c(abc = 123)[1].

The index object i can be numeric, logical, character or empty. Indexing by factors is allowed and is equivalent to indexing by the numeric codes (see factor) and not by the character values which are printed (for which use [as.character(i)]).

An empty index selects all values: this is most often used to replace all the entries but keep the attributes.

Matrices and arrays

Matrices and arrays are vectors with a dimension attribute and so all the vector forms of indexing can be used with a single index. The result will be an unnamed vector unless x is one-dimensional when it will be a one-dimensional array.

The most common form of indexing a \(k\)-dimensional array is to specify \(k\) indices to [. As for vector indexing, the indices can be numeric, logical, character, empty or even factor. An empty index (a comma separated blank) indicates that all entries in that dimension are selected. The argument drop applies to this form of indexing.

A third form of indexing is via a numeric matrix with the one column for each dimension: each row of the index matrix then selects a single element of the array, and the result is a vector. Negative indices are not allowed in the index matrix. NA and zero values are allowed: rows of an index matrix containing a zero are ignored, whereas rows containing an NA produce an NA in the result.

Indexing via a character matrix with one column per dimensions is also supported if the array has dimension names. As with numeric matrix indexing, each row of the index matrix selects a single element of the array. Indices are matched against the appropriate dimension names. NA is allowed and will produce an NA in the result. Unmatched indices as well as the empty string ("") are not allowed and will result in an error.

A vector obtained by matrix indexing will be unnamed unless x is one-dimensional when the row names (if any) will be indexed to provide names for the result.

Recursive (list-like) objects

Indexing by [ is similar to atomic vectors and selects a list of the specified element(s).

Both [[ and $ select a single element of the list. The main difference is that $ does not allow computed indices, whereas [[ does. x$name is equivalent to x[["name", exact = FALSE]]. Also, the partial matching behavior of [[ can be controlled using the exact argument.

getElement(x, name) is a version of x[[name, exact = TRUE]] which for formally classed (S4) objects returns slot(x, name), hence providing access to even more general list-like objects.

[ and [[ are sometimes applied to other recursive objects such as calls and expressions. Pairlists are coerced to lists for extraction by [, but all three operators can be used for replacement.

[[ can be applied recursively to lists, so that if the single index i is a vector of length p, alist[[i]] is equivalent to alist[[i1]]…[[ip]] providing all but the final indexing results in a list.

Note that in all three kinds of replacement, a value of NULL deletes the corresponding item of the list. To set entries to NULL, you need x[i] <- list(NULL).

When $<- is applied to a NULL x, it first coerces x to list(). This is what also happens with [[<- if the replacement value value is of length greater than one: if value has length 1 or 0, x is first coerced to a zero-length vector of the type of value.


Both $ and [[ can be applied to environments. Only character indices are allowed and no partial matching is done. The semantics of these operations are those of get(i, env = x, inherits = FALSE). If no match is found then NULL is returned. The replacement versions, $<- and [[<-, can also be used. Again, only character arguments are allowed. The semantics in this case are those of assign(i, value, env = x, inherits = FALSE). Such an assignment will either create a new binding or change the existing binding in x.

NAs in indexing

When extracting, a numerical, logical or character NA index picks an unknown element and so returns NA in the corresponding element of a logical, integer, numeric, complex or character result, and NULL for a list. (It returns 00 for a raw result.)

When replacing (that is using indexing on the lhs of an assignment) NA does not select any element to be replaced. As there is ambiguity as to whether an element of the rhs should be used or not, this is only allowed if the rhs value is of length one (so the two interpretations would have the same outcome). (The documented behaviour of S was that an NA replacement index ‘goes nowhere’ but uses up an element of value: Becker et al p.359. However, that has not been true of other implementations.)

Argument matching

Note that these operations do not match their index arguments in the standard way: argument names are ignored and positional matching only is used. So m[j = 2, i = 1] is equivalent to m[2, 1] and not to m[1, 2].

This may not be true for methods defined for them; for example it is not true for the data.frame methods described in [.data.frame which warn if i or j is named and have undocumented behaviour in that case.

To avoid confusion, do not name index arguments (but drop and exact must be named).

S4 methods

These operators are also implicit S4 generics, but as primitives, S4 methods will be dispatched only on S4 objects x.

The implicit generics for the $ and $<- operators do not have name in their signature because the grammar only allows symbols or string constants for the name argument.

Character indices

Character indices can in some circumstances be partially matched (see pmatch) to the names or dimnames of the object being subsetted (but never for subassignment). Unlike S (Becker et al p.358)), R never uses partial matching when extracting by [, and partial matching is not by default used by [[ (see argument exact).

Thus the default behaviour is to use partial matching only when extracting from recursive objects (except environments) by $. Even in that case, warnings can be switched on by options(warnPartialMatchDollar = TRUE).

Neither empty ("") nor NA indices match any names, not even empty nor missing names. If any object has no names or appropriate dimnames, they are taken as all "" and so match nothing.


These operators are generic. You can write methods to handle indexing of specific classes of objects, see InternalMethods as well as [.data.frame and [.factor. The descriptions here apply only to the default methods. Note that separate methods are required for the replacement functions [<-, [[<- and $<- for use when indexing occurs on the assignment side of an expression.

The most important distinction between [, [[ and $ is that the [ can select more than one element whereas the other two select a single element.

The default methods work somewhat differently for atomic vectors, matrices/arrays and for recursive (list-like, see is.recursive) objects. $ is only valid for recursive objects, and is only discussed in the section below on recursive objects.

Subsetting (except by an empty index) will drop all attributes except names, dim and dimnames.

Indexing can occur on the right-hand-side of an expression for extraction, or on the left-hand-side for replacement. When an index expression appears on the left side of an assignment (known as subassignment) then that part of x is set to the value of the right hand side of the assignment. In this case no partial matching of character indices is done, and the left-hand-side is coerced as needed to accept the values. For vectors, the answer will be of the higher of the types of x and value in the hierarchy raw < logical < integer < double < complex < character < list < expression. Attributes are preserved (although names, dim and dimnames will be adjusted suitably). Subassignment is done sequentially, so if an index is specified more than once the latest assigned value for an index will result.

It is an error to apply any of these operators to an object which is not subsettable (e.g., a function).


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

See Also

names for details of matching to names, and pmatch for partial matching.

list, array, matrix.

[.data.frame and [.factor for the behaviour when applied to data.frame and factors.

Syntax for operator precedence, and the ‘R Language Definition’ manual about indexing details.

NULL for details of indexing null objects.


Run this code
x <- 1:12
m <- matrix(1:6, nrow = 2, dimnames = list(c("a", "b"), LETTERS[1:3]))
li <- list(pi = pi, e = exp(1))
x[10]                 # the tenth element of x
x <- x[-1]            # delete the 1st element of x
m[1,]                 # the first row of matrix m
m[1, , drop = FALSE]  # is a 1-row matrix
m[,c(TRUE,FALSE,TRUE)]# logical indexing
m[cbind(c(1,2,1),3:1)]# matrix numeric index
ci <- cbind(c("a", "b", "a"), c("A", "C", "B"))
m[ci]                 # matrix character index
m <- m[,-1]           # delete the first column of m
li[[1]]               # the first element of list li
y <- list(1, 2, a = 4, 5)
y[c(3, 4)]            # a list containing elements 3 and 4 of y
y$a                   # the element of y named a

## non-integer indices are truncated:
(i <- 3.999999999) # "4" is printed
(1:5)[i]  # 3

## named atomic vectors, compare "[" and "[[" :
nx <- c(Abc = 123, pi = pi)
nx[1] ; nx["pi"] # keeps names, whereas "[[" does not:
nx[[1]] ; nx[["pi"]]

## recursive indexing into lists
z <- list(a = list(b = 9, c = "hello"), d = 1:5)
z[[c(1, 2)]]
z[[c(1, 2, 1)]]  # both "hello"
z[[c("a", "b")]] <- "new"

## check $ and [[ for environments
e1 <- new.env()
e1$a <- 10
e1[["b"]] <- 20

## partial matching - possibly with warning :
stopifnot(identical(li$p, pi))
op <- options(warnPartialMatchDollar = TRUE)
stopifnot( identical(li$p, pi), #-- a warning
  inherits(tryCatch (li$p, warning = identity), "warning"))
## revert the warning option:
if(is.null(op[[1]])) op[[1]] <- FALSE; options(op)

Run the code above in your browser using DataLab