In Nelson-Gon/mde: Missing Data Explorer

knitr::opts_chunk$set(
  collapse = TRUE,
  comment = "#>"
)

The goal of mde is to ease exploration of missingness.

Installation

CRAN release

install.packages("mde")

Stable Development version

devtools::install_github("Nelson-Gon/mde")


devtools::install_github("Nelson-Gon/mde",  build_vignettes=TRUE)

Unstable Development version

devtools::install_github("Nelson-Gon/mde@develop")

Loading the package

library(mde)

Exploring missingness

To get a simple missingness report, use na_summary:

na_summary(airquality)

To sort this summary by a given column :

na_summary(airquality,sort_by = "percent_complete")

If one would like to reset (drop) row names, then one can set row_names to TRUE This may especially be useful in cases where rownames are simply numeric and do not have much additional use.

na_summary(airquality,sort_by = "percent_complete", reset_rownames = TRUE)

To sort by percent_missing instead:

na_summary(airquality, sort_by = "percent_missing")

To sort the above in descending order:

na_summary(airquality, sort_by="percent_missing", descending = TRUE)

To exclude certain columns from the analysis:

na_summary(airquality, exclude_cols = c("Day", "Wind"))

To include or exclude via regex match:

na_summary(airquality, regex_kind = "inclusion",pattern_type = "starts_with", pattern = "O|S")

na_summary(airquality, regex_kind = "exclusion",pattern_type = "regex", pattern = "^[O|S]")

To get this summary by group:

test2 <- data.frame(ID= c("A","A","B","A","B"), Vals = c(rep(NA,4),"No"),ID2 = c("E","E","D","E","D"))

na_summary(test2,grouping_cols = c("ID","ID2"))

na_summary(test2, grouping_cols="ID")

get_na_counts

This provides a convenient way to show the number of missing values column-wise. It is relatively fast(tests done on about 400,000 rows, took a few microseconds.)

To get the number of missing values in each column of airquality, we can use the function as follows:

get_na_counts(airquality)

The above might be less useful if one would like to get the results by group. In that case, one can provide a grouping vector of names in grouping_cols.

test <- structure(list(Subject = structure(c(1L, 1L, 2L, 2L), .Label = c("A", 
"B"), class = "factor"), res = c(NA, 1, 2, 3), ID = structure(c(1L, 
1L, 2L, 2L), .Label = c("1", "2"), class = "factor")), class = "data.frame", row.names = c(NA, 
-4L))

get_na_counts(test, grouping_cols = "ID")

percent_missing

This is a very simple to use but quick way to take a look at the percentage of data that is missing column-wise.

percent_missing(airquality)

We can get the results by group by providing an optional grouping_cols character vector.

percent_missing(test, grouping_cols = "Subject")

To exclude some columns from the above exploration, one can provide an optional character vector in exclude_cols.

percent_missing(airquality,exclude_cols = c("Day","Temp"))

sort_by_missingness

This provides a very simple but relatively fast way to sort variables by missingness. Unless otherwise stated, this does not currently support arranging grouped percents.

Usage:

sort_by_missingness(airquality, sort_by = "counts")

To sort in descending order:

sort_by_missingness(airquality, sort_by = "counts", descend = TRUE)

To use percentages instead:

sort_by_missingness(airquality, sort_by = "percents")

Recoding as NA

recode_as_na

As the name might imply, this converts any value or vector of values to NA i.e. we take a value such as "missing" or "NA" (not a real NA according to R) and convert it to R's known handler for missing values (NA).

To use the function out of the box (with default arguments), one simply does something like:

dummy_test <- data.frame(ID = c("A","B","B","A"), 
                         values = c("n/a",NA,"Yes","No"))
# Convert n/a and no to NA
head(recode_as_na(dummy_test, value = c("n/a","No")))

Great, but I want to do so for specific columns not the entire dataset. You can do this by providing column names to subset_cols.

another_dummy <- data.frame(ID = 1:5, Subject = 7:11, 
Change = c("missing","n/a",2:4 ))
# Only change values at the column Change
head(recode_as_na(another_dummy, subset_cols = "Change", value = c("n/a","missing")))

To recode columns using RegEx,one can provide pattern_type and a target pattern. Currently supported pattern_types are starts_with, ends_with, contains and regex. See docs for more details.:

# only change at columns that start with Solar
head(recode_as_na(airquality,value=190,pattern_type="starts_with",pattern="Solar"))

# recode at columns that start with O or S(case sensitive)
head(recode_as_na(airquality,value=c(67,118),pattern_type="starts_with",pattern="S|O"))

# use my own RegEx
head(recode_as_na(airquality,value=c(67,118),pattern_type="regex",pattern="(?i)^(s|o)"))

recode_as_na_if

This function allows one to deliberately introduce missing values if a column meets a certain threshold of missing values. This is similar to amputation but is much more basic. It is only provided here because it is hoped it may be useful to someone for whatever reason.

head(recode_as_na_if(airquality,sign="gt", percent_na=20))

recode_as_na_str

This allows recoding as NA based on a string match.

partial_match <- data.frame(A=c("Hi","match_me","nope"), B=c(NA, "not_me","nah"))

recode_as_na_str(partial_match,"ends_with","ME", case_sensitive=FALSE)

recode_as_na_for

For all values greater/less/less or equal/greater or equal than some value, can I convert them to NA?!

Yes You Can! All we have to do is use recode_as_na_for:

head(recode_as_na_for(airquality,criteria="gt",value=25))

To do so at specific columns, pass an optional subset_cols character vector:

head(recode_as_na_for(airquality, value=40,subset_cols=c("Solar.R","Ozone"), criteria="gt"))

Recoding NA as

recode_na_as

Sometimes, for whatever reason, one would like to replace NAs with whatever value they would like. recode_na_as provides a very simple way to do just that.

head(recode_na_as(airquality))

# use NaN

head(recode_na_as(airquality, value=NaN))

As a "bonus", you can manipulate the data only at specific columns as shown here:

head(recode_na_as(airquality, value=0, subset_cols="Ozone"))

The above also supports custom recoding similar to recode_na_as:

head(mde::recode_na_as(airquality, value=0, pattern_type="starts_with",pattern="Solar"))

column_based_recode

Ever needed to change values in a given column based on the proportions of NAs in other columns(row-wise)?!. The goal of column_based_recode is to achieve just that. Let's see how we could do this with a simple example:

head(column_based_recode(airquality, values_from = "Wind", values_to="Wind", pattern_type = "regex", pattern = "Solar|Ozone"))

custom_na_recode

This allows recoding NA values with common stats functions such as mean,max,min,sd.

To use default values:

head(custom_na_recode(airquality))

To use select columns:

head(custom_na_recode(airquality,func="mean",across_columns=c("Solar.R","Ozone")))

To use a function from another package to perform replacements:

To perform a forward fill with dplyr's lead:

# use lag for a backfill
head(custom_na_recode(airquality,func=dplyr::lead ))

To perform replacement by group:

some_data <- data.frame(ID=c("A1","A1","A1","A2","A2", "A2"),A=c(5,NA,0,8,3,4),B=c(10,0,0,NA,5,6),C=c(1,NA,NA,25,7,8))

head(custom_na_recode(some_data,func = "mean", grouping_cols = "ID"))

Across specific columns:

head(custom_na_recode(some_data,func = "mean", grouping_cols = "ID", across_columns = c("C", "A")))

recode_na_if

Given a data.frame object, one can recode NAs as another value based on a grouping variable. In the example below, we replace all NAs in all columns with 0s if the ID is A2 or A3

some_data <- data.frame(ID=c("A1","A2","A3", "A4"), 
                        A=c(5,NA,0,8), B=c(10,0,0,1),
                        C=c(1,NA,NA,25))

head(recode_na_if(some_data,grouping_col="ID", target_groups=c("A2","A3"),
           replacement= 0))

Dropping NAs

drop_na_if

Suppose you wanted to drop any column that has a percentage of NAs greater than or equal to a certain value? drop_na_if does just that.

We can drop any columns that have greater than or equal(gteq) to 24% of the values missing from airquality:

head(drop_na_if(airquality, sign="gteq",percent_na = 24))

The above also supports less than or equal to(lteq), equal to(eq), greater than(gt) and less than(lt).

To keep certain columns despite fitting the target percent_na criteria, one can provide an optional keep_columns character vector.

head(drop_na_if(airquality, percent_na = 24, keep_columns = "Ozone"))

Compare the above result to the following:

head(drop_na_if(airquality, percent_na = 24))

To drop groups that meet a set missingness criterion, we proceed as follows.

grouped_drop <- structure(list(ID = c("A", "A", "B", "A", "B"), 
          Vals = c(4, NA,  NA, NA, NA), Values = c(5, 6, 7, 8, NA)), 
          row.names = c(NA, -5L), class = "data.frame")
# Drop all columns for groups that meet a percent missingness of greater than or
# equal to 67
drop_na_if(grouped_drop,percent_na = 67,sign="gteq",
                                    grouping_cols = "ID")

drop_row_if

This is similar to drop_na_if but does operations rowwise not columnwise. Compare to the example above:

# Drop rows with at least two NAs
head(drop_row_if(airquality, sign="gteq", type="count" , value = 2))

To drop based on percentages:

# Drops 42 rows
head(drop_row_if(airquality, type="percent", value=16, sign="gteq",
                 as_percent=TRUE))

For more details, please see the documentation of drop_row_if.

drop_na_at

This provides a simple way to drop missing values only at specific columns. It currently only returns those columns with their missing values removed. See usage below. Further details are given in the documentation. It is currently case sensitive.

head(drop_na_at(airquality,pattern_type = "starts_with","O"))

drop_all_na

This drops columns where all values are missing.

test2 <- data.frame(ID= c("A","A","B","A","B"), Vals = c(4,rep(NA, 4))) 
drop_all_na(test2, grouping_cols="ID")

Alternatively, we can drop groups where all variables are all NA.

test2 <- data.frame(ID= c("A","A","B","A","B"), Vals = rep(NA, 5)) 

head(drop_all_na(test, grouping_cols = "ID"))

dict_recode

If one would like to recode column values using a "dictionary", dict_recode provides a simple way to do that. For example, if one would like to convert NA values in Solar.R to 520 and those in Ozone to 42, one simply calls the following:

head(dict_recode(airquality, use_func="recode_na_as",
                 patterns = c("solar", "ozone"),
                 pattern_type="starts_with", values = c(520,42)))