Impute Missing Values in a Vector

Replaces missing values (NA) in a vector by either a summary statistic computed from the data (e.g. mean, median) or a user-supplied scalar value.

Usage

impute(x, FUN = median, na.rm = TRUE, ...)

Arguments

x

atomic vector

FUN

function used to compute the imputation value, or a scalar value; default is median

na.rm

logical; whether to remove missing values before computing the summary statistic

...

additional arguments passed to FUN

Value

vector of the same length as x, with missing values replaced. The type follows R's usual coercion rules: imputing a double (the median of an integer vector, say) into an integer vector returns a double.

Details

If FUN is a function, it is applied to x to compute a single imputation value. If the function accepts an argument named na.rm, it will be passed automatically.

Alternatively, FUN can be a single scalar value, which will directly replace all missing values.

If FUN is a function, it must return a single value. An error is thrown if the returned value is not scalar.

Whether FUN accepts na.rm is decided from its formal arguments; a function taking \dots is assumed to pass it on. Functions without either are called without it.

See also

For direct value replacement see bedrock::naReplace().

Other impute: imputeKnn()

Examples

x <- c(2, 3, NA, 5, 9)

# Default: median(x, na.rm=TRUE)
impute(x)
#> [1] 2 3 4 5 9

# Using mean(x, na.rm=TRUE)
impute(x, mean)
#> [1] 2.00 3.00 4.75 5.00 9.00

# Using trimmed mean
impute(x, function(x) meanX(x, trim=0.3, na.rm = TRUE))
#> [1] 2 3 4 5 9

# Constant replacement
impute(x, 99)
#> [1]  2  3 99  5  9

# an integer vector stays integer only if the value is one too
impute(c(2L, 3L, NA, 5L), FUN = 4L)
#> [1] 2 3 4 5