Add Missing Values Completely at Random

Sets a fixed share of cells in each selected column to NA, choosing the rows independently for each column (missing completely at random). Use it to test imputation: start from complete data, add missingness, impute, and compare the imputed cells with the originals.

Usage

add_missing(data, cols = dplyr::everything(), pct_na = 0.1)

Arguments

data

A data frame or tibble.

cols

Columns to add missing values to, as a tidyselect expression (default dplyr::everything()).

pct_na

Proportion of cells to set to NA in each selected column, between 0 and 1 (default 0.1). Can be a single number applied to all selected columns, or a numeric vector with length equal to the number of selected columns specifying column-specific amputation rates.

Value

data with missing values added. Its "missing_cells" attribute is a tibble with one row per masked cell, giving its row number and feature (column name). Most dplyr verbs drop this attribute, so read it before transforming the result.

Details

Every selected column must start with no missing values, so afterwards its share of NAs equals pct_na (to the nearest whole row), matching PctNa from features_percent_miss. The observed share is 1 - pct_na, so only pct_na is given.

Examples

set.seed(1)
amputed <- add_missing(mtcars, c(mpg, hp), pct_na = 0.25)
features_percent_miss(amputed)
#> # A tibble: 11 × 5
#>    feature SumNa SumComp PctNa PctComp
#>    <chr>   <int>   <int> <dbl>   <dbl>
#>  1 mpg         8      24  0.25    0.75
#>  2 hp          8      24  0.25    0.75
#>  3 cyl         0      32  0       1   
#>  4 disp        0      32  0       1   
#>  5 drat        0      32  0       1   
#>  6 wt          0      32  0       1   
#>  7 qsec        0      32  0       1   
#>  8 vs          0      32  0       1   
#>  9 am          0      32  0       1   
#> 10 gear        0      32  0       1   
#> 11 carb        0      32  0       1   
head(attr(amputed, "missing_cells"))
#> # A tibble: 6 × 2
#>     row feature
#>   <int> <chr>  
#> 1     1 mpg    
#> 2     2 mpg    
#> 3     4 mpg    
#> 4     7 mpg    
#> 5    11 mpg    
#> 6    14 mpg    

# Column-specific missingness proportions with a vector:
p_vec <- c(mpg = 0.1, hp = 0.3)
amputed2 <- add_missing(mtcars, c(mpg, hp), pct_na = p_vec)
features_percent_miss(amputed2)
#> # A tibble: 11 × 5
#>    feature SumNa SumComp  PctNa PctComp
#>    <chr>   <int>   <int>  <dbl>   <dbl>
#>  1 hp         10      22 0.312    0.688
#>  2 mpg         3      29 0.0938   0.906
#>  3 cyl         0      32 0        1    
#>  4 disp        0      32 0        1    
#>  5 drat        0      32 0        1    
#>  6 wt          0      32 0        1    
#>  7 qsec        0      32 0        1    
#>  8 vs          0      32 0        1    
#>  9 am          0      32 0        1    
#> 10 gear        0      32 0        1    
#> 11 carb        0      32 0        1