impute {Hmisc}

Generic Functions and Methods for Imputation

Description

These functions do simple and transcan imputation and print, summarize, and subscript variables that have NAs filled-in with imputed values. The simple imputation method involves filling in NAs with constants, with a specified single-valued function of the non-NAs, or from a sample (with replacement) from the non-NA values (this is useful in multiple imputation). More complex imputations can be done with the transcan function, which also works with the generic methods shown here, i.e., impute can take a transcan object and use the imputed values created by transcan (with imputed=TRUE) to fill-in NAs. The print method places * after variable values that were imputed. The summary method summarizes all imputed values and then uses the next summary method available for the variable. The subscript method preserves attributes of the variable and subsets the list of imputed values corresponding with how the variable was subsetted. The is.imputed function is for checking if observations are imputed.

Usage

impute(x, ...)

## Default S3 method:
impute(x, fun=median, ...)

## S3 method for class 'impute'
print(x, ...)

## S3 method for class 'impute'
summary(object, ...)

is.imputed(x)

Arguments

x

a vector or an object created by transcan, or a vector needing basic unconditional imputation. If there are no NAs and x is a vector, it is returned unchanged.

fun

the name of a function to use in computing the (single) imputed value from the non-NAs. The default is median. If instead of specifying a function as fun, a single value or vector (numeric, or character if object is a factor) is specified, those values are used for insertion. fun can also be the character string "random" to draw random values for imputation, with the random values not forced to be the same if there are multiple NAs. For a vector of constants, the vector must be of length one (indicating the same value replaces all NAs) or must be as long as the number of NAs, in which case the values correspond to consecutive NAs to replace. For a factor object, constants for imputation may include character values not in the current levels of object. In that case new levels are added. If object is of class "factor", fun is ignored and the most frequent category is used for imputation.

object

an object of class "impute"

...

ignored

Value

a vector with class "impute" placed in front of existing classes. For is.imputed, a vector of logical values is returned (all TRUE if object is not of class impute).

Author(s)

Frank Harrell
Department of Biostatistics
Vanderbilt University
[e-mail]

See Also

transcan, impute.transcan, describe, na.include, sample

Examples

age <- c(1,2,NA,4)
age.i <- impute(age)
# Could have used impute(age,2.5), impute(age,mean), impute(age,"random")
age.i
summary(age.i)
is.imputed(age.i)