Announcing the simputation package: make imputation simple

(This article was first published on R – Mark van der Loo, and kindly contributed to R-bloggers)

I am happy to announce that my simputation package has appeared on CRAN this weekend. This package aims to simplify missing value imputation. In particular it offers standardized interfaces that

  • make it easy to define both imputation method and imputation model;
  • for multiple variables at once;
  • while grouping data by categorical variables;
  • all fitting in the magrittr not-a-pipeline.

A few examples

To start with an example, let us first create a data set with some missings.

dat <- iris
# empty a few fields
dat[1:3,1] <- dat[3:7,2] <- dat[8:10,5] <- NA
head(dat,10)
##    Sepal.Length Sepal.Width Petal.Length Petal.Width Species
## 1            NA         3.5          1.4         0.2  setosa
## 2            NA         3.0          1.4         0.2  setosa
## 3            NA          NA          1.3         0.2  setosa
## 4           4.6          NA          1.5         0.2  setosa
## 5           5.0          NA          1.4         0.2  setosa
## 6           5.4          NA          1.7         0.4  setosa
## 7           4.6          NA          1.4         0.3  setosa
## 8           5.0         3.4          1.5         0.2    <NA>
## 9           4.4         2.9          1.4         0.2    <NA>
## 10          4.9         3.1          1.5         0.1    <NA>

Below, we first impute Sepal.Width and Sepal.Length by regression on Petal.Width and Species. After this we impute Species using a decision tree model (CART) using every other variable as a predictor (including the ones just imputed).

library(magrittr)    # load the %>% operator
library(simputation) 
imputed <- dat %>% 
  impute_lm(Sepal.Width + Sepal.Length ~ Petal.Width + Species) %>%
  impute_cart(Species ~ .)
head(imputed,10)
##    Sepal.Length Sepal.Width Petal.Length Petal.Width Species
## 1      4.979844    3.500000          1.4         0.2  setosa
## 2      4.979844    3.000000          1.4         0.2  setosa
## 3      4.979844    3.409547          1.3         0.2  setosa
## 4      4.600000    3.409547          1.5         0.2  setosa
## 5      5.000000    3.409547          1.4         0.2  setosa
## 6      5.400000    3.561835          1.7         0.4  setosa
## 7      4.600000    3.485691          1.4         0.3  setosa
## 8      5.000000    3.400000          1.5         0.2  setosa
## 9      4.400000    2.900000          1.4         0.2  setosa
## 10     4.900000    3.100000          1.5         0.1  setosa

The package is pretty lenient against failure of imputation. For example, if one of the predictors is missing, fields just remain unimputed and if one of the models cannot be fitted, only a warning is issued (not shown here).

dat %>% impute_lm(Sepal.Length ~ Sepal.Width + Species) %>% head(3)
##   Sepal.Length Sepal.Width Petal.Length Petal.Width Species
## 1     5.076579         3.5          1.4         0.2  setosa
## 2     4.675654         3.0          1.4         0.2  setosa
## 3           NA          NA          1.3         0.2  setosa

So here, the third Sepal.Length value could not be imputed since the predictor Sepal.Width is missing.

It is possible to split data into groups before estimating the imputation model and predicting missing values. There are two ways. The first is to use the | operator to specify grouping variables.

# We first need to complete 'Species'. Here, we use sequential 
# hot deck after sorting by Petal.Length
dat %<>% impute_shd(Species ~ Petal.Length) 
# Now impute Sepal.Length by regressing on 
# Sepal.Width, computing a model for each Species.
dat %>% impute_lm(Sepal.Length ~ Sepal.Width | Species) %>% head(3)
##   Sepal.Length Sepal.Width Petal.Length Petal.Width Species
## 1     5.067813         3.5          1.4         0.2  setosa
## 2     4.725677         3.0          1.4         0.2  setosa
## 3           NA          NA          1.3         0.2  setosa

The second way is to use the group_by command from dplyr

dat %>% dplyr::group_by(Species) %>% 
    impute_lm(Sepal.Length ~ Sepal.Width) %>% 
    head(3)
## Source: local data frame [3 x 5]
## Groups: Species [1]
## 
##   Sepal.Length Sepal.Width Petal.Length Petal.Width Species
##          <dbl>       <dbl>        <dbl>       <dbl>  <fctr>
## 1     5.067813         3.5          1.4         0.2  setosa
## 2     4.725677         3.0          1.4         0.2  setosa
## 3           NA          NA          1.3         0.2  setosa

Note: by using group_by, we also transformed the data.frame to a tibble, which not only sounds funny when you pronounce it (tibble, TIBBLE, tibble? tibbebbebbebble) but is also pretty useful.

Supported methods and how to specify them

Currently, the package supports the following methods:

  • Model based (optionally add [non-]parametric random residual)
    • linear regression
    • robust linear regression
    • CART models
    • Random forest
  • Donor imputation (including various donor pool specifications)
    • k-nearest neigbour (based on gower‘s distance)
    • sequential hotdeck (LOCF, NOCB)
    • random hotdeck
    • Predictive mean matching
  • Other
    • (groupwise) median imputation (optional random residual)
    • Proxy imputation (copy from other variable)

Any call to one of the impute_ functions looks as follows:

impute_<method>(data, formula [, <method-specific options>])

and the formula always has the following form:

<imputed variables> ~ <model specification> [|<grouping variables>]

The parts in square brackets are optional.

Please see the package vignette for more examples and details, or ?simputation::impute_ for an overview of all imputation functions.

Happy imputing!