R: Take a Stratified Sample From a Dataset

stratified {splitstackshape}

R Documentation

Take a Stratified Sample From a Dataset

Description

The stratified function samples from a data.frame or a data.table in which one or more columns can be used as a "stratification" or "grouping" variable. The result is a new data.table with the specified number of samples from each group.

Usage

stratified(indt, group, size, select = NULL, replace = FALSE,
  keep.rownames = FALSE, bothSets = FALSE, ...)

Arguments

`indt`	The input `data.frame` or `data.table`.
`group`	The column or columns that should be used to create the groups. Can be a character vector of column names (recommended) or a numeric vector of column positions. Generally, if you are using more than one variable to create your "strata", you should list them in the order of slowest varying to quickest varying. This can be a vector of names or column indexes.
`size`	The desired sample size. If `size` is a value between `0` and `1` expressed as a decimal, size is set to be proportional to the number of observations per group. If `size` is a single positive integer, it will be assumed that you want the same number of samples from each group. If `size` is a named vector, the function will check to see whether the length of the vector matches the number of groups and that the names match the group names.
`select`	A named list containing levels from the "group" variables in which you are interested. The list names must be present as variable names for the input dataset.
`replace`	Logical. Should sampling be with replacement? Defaults to `FALSE`.
`keep.rownames`	Logical. If the input is a `data.frame` or a `matrix`, `as.data.table` would normally drop the rownames. If `TRUE`, the rownames would be retained in a column named `rn`. Defaults to `FALSE`.
`bothSets`	Logical. Should both the sampled and non-sampled sets be returned as a `list`? Defaults to `FALSE`.
`...`	Optional arguments to `sample`.

Value

If bothSets = FALSE, a list of two data.tables; otherwise, a data.table.

Note

Slightly different sizes than requested: Because of how computers deal with floating-point arithmetic, and because R uses a "round to even" approach, the size per strata that results when specifying a proportionate sample may be slightly higher or lower per strata than you might have expected.

Author(s)

Ananda Mahto

Examples

# Generate a sample data.frame to play with
set.seed(1)
dat1 <- data.frame(ID = 1:100,
              A = sample(c("AA", "BB", "CC", "DD", "EE"),
                         100, replace = TRUE),
              B = rnorm(100), C = abs(round(rnorm(100), digits=1)),
              D = sample(c("CA", "NY", "TX"), 100, replace = TRUE),
              E = sample(c("M", "F"), 100, replace = TRUE))

# Let's take a 10% sample from all -A- groups in dat1
stratified(dat1, "A", .1)

# Let's take a 10% sample from only "AA" and "BB" groups from -A- in dat1
stratified(dat1, "A", .1, select = list(A = c("AA", "BB")))

# Let's take 5 samples from all -D- groups in dat1,
#   specified by column number
stratified(dat1, group = 5, size = 5)

# Use a two-column strata: -E- and -D-
#   -E- varies more slowly, so it is better to put that first
stratified(dat1, c("E", "D"), size = .15)

# Use a two-column strata (-E- and -D-) but only interested in
#   cases where -E- == "M"
stratified(dat1, c("E", "D"), .15, select = list(E = "M"))

## As above, but where -E- == "M" and -D- == "CA" or "TX"
stratified(dat1, c("E", "D"), .15,
     select = list(E = "M", D = c("CA", "TX")))

# Use a three-column strata: -E-, -D-, and -A-
s.out <- stratified(dat1, c("E", "D", "A"), size = 2)

[Package splitstackshape version 1.4.2 Index]