# naniar 0.6.0.9000 Unreleased

## New features

• naniar now provides mcar_test() for Little’s (1988) statistical test for missing completely at random (MCAR) data. The null hypothesis in this test is that the data is MCAR, and the test statistic is a chi-squared value. Given a high statistic value and low p-value, we can conclude data are not missing completely at random. Thanks to Andrew Heiss for the PR.

## Bug fixes

• Fix bug in miss_var_span() (#270) where the number of missings + number of complete values added up to more than the number of rows in the data. This was due to the remainder not being used when calculating the number of complete values.
• Fix bug in recode_shadow() (#272) where adding the same special missing value in two subsequent operations fails.

# naniar 0.6.0 (2020/08/17) “Spur of the lamp post” 2020-09-02

• Provide warning for replace_with_na when columns provided that don’t exist (see #160). Thank you to michael-dewar for their help with this.

## Breaking Changes

• Drop the “nabular” and “shadow” classes (#268) used in nabular() and bind_shadow(). In doing so removes the functions, as_shadow(), is_shadow(), is_nabular(), new_nabular(), new_shadow(). These were mostly used internally and it is not expected that users would have used this functions. If these were used, please file an issue and I can implement them again.

# naniar 0.5.2 (2020/06/28) “Silver Apple” 2020-06-29

## Minor Changes

• Improvements to code in miss_var_summary(), miss_var_table(), and prop_miss_var(), resulting in a 3-10x speedup.

# naniar 0.5.1 (2020/04/10) “Uncle Andrew’s Applewood Wardrobe” 2020-04-30

## Minor Changes

• Fixes warnings and errors from tibble and subsequent downstream impacts on simputation.

# naniar 0.5.0 (2020/02/20) “The End of this Story and the Beginning of all of the Others” 2020-02-28

## Breaking Changes

• The following functions related to calculating the proportion/percentage of missingness were made Defunct and will no longer work:
• miss_var_prop()
• complete_var_prop()
• miss_var_pct()
• complete_var_pct()
• miss_case_prop()
• complete_case_prop()
• miss_case_pct()
• complete_case_pct()

Instead use: prop_miss_var(), prop_complete_var(), pct_miss_var(), pct_complete_var(), prop_miss_case(), prop_complete_case(), pct_miss_case(), pct_complete_case(). (see 242)

• replace_to_na() was made defunct, please use replace_with_na() instead. (see 242)

## Minor changes

• miss_var_cumsum and miss_case_cumsum are now exported
• use map_dfc instead of map_df
• Fix various extra warnings and improve test coverage

## Bug Fixes

• Address bug where the number of missings in a row is not calculated properly - see 238 and 232. The solution involved using rowSums(is.na(x)), which was 3 times faster.
• Resolve bug in gg_miss_fct() where warning is given for non explicit NA values - see 241.
• skip vdiffr tests on github actions
• use tibble() not data_frame()

# naniar 0.4.2 (2019/02/15) “The Planting of The Tree” 2019-02-15

## Improvements

• The geom_miss_point() ggplot2 layer can now be converted into an interactive web-based version by the ggplotly() function in the plotly package. In order for this to work, naniar now exports the geom2trace.GeomMissPoint() function (users should never need to call geom2trace.GeomMissPoint() directly – ggplotly() calls it for you).
• adds WORDLIST for spelling thanks to usethis::use_spell_check()
• fix documentation @seealso bug (#228) (@sfirke)

## Dependency fixes

• Thanks to a PR (#223) from @romainfrancois:

• This fixes two problems that were identified as part of reverse dependency checks of dplyr 0.8.0 release candidate. https://github.com/tidyverse/dplyr/blob/revdep_dplyr_0_8_0_RC/revdep/problems.md#naniar

• n() must be imported or prefixed like any other function. In the PR, I’ve changed 1:n() to dplyr::row_number() as naniar seems to prefix all dplyr functions.

• update_shadow was only restoring the class attributes, changed so that it restores all attributes, this was causing problems when data was a grouped_df. This likely was a problem before too, but dplyr 0.8.0 is stricter about what is a grouped data frame.

# naniar 0.4.1 (2018/12/14) 2018-11-20

## Minor Changes

• pkgdown updates: update favicon and logo, set up for gh-pages deployment
• use a scalar integer in new_tibble

# naniar 0.4.0 (2018/09/10) “An Unexpected Meeting” Unreleased

## New Feature

• Add custom label support for missings and not missings with functions add_label_missings and add_label_shadow() and add_any_miss(). So you can now do add_label_missings(data, missing = “custom_missing_label”, complete = “custom_complete_label”)

• impute_median() and scoped variants

• any_shade() returns a logical TRUE or FALSE depending on if there are any shade values

• nabular() an alias for bind_shadow() to tie the nabular term into the work.

• is_nabular() checks if input is nabular.

• geom_miss_point() now gains the arguments from shadow_shift()/impute_below() for altering the amount of jitter and proportion below (prop_below).

• Added two new vignettes, “Exploring Imputed Values”, and “Special Missing Values”

• miss_var_summary and miss_case_summary now no longer provide the cumulative sum of missingness in the summaries - this summary can be added back to the data with the option add_cumsum = TRUE. #186

• Added gg_miss_upset to replace workflow of:

data %>%
UpSetR::upset()

## Major Change

• recode_shadow now works! This function allows you to recode your missing values into special missing values. These special missing values are stored in the shadow part of the dataframe, which ends in _NA.
• implemented shade where appropriate throughout naniar, and also added verifiers, is_shade, are_shade, which_are_shade, and removed which_are_shadow.
• as_shadow and bind_shadow now return data of class shadow. This will feed into recode_shadow methods for flexibly adding new types of missing data.
• Note that in the future shadow might be changed to nabble or something similar.

## Minor feature

• Functions add_label_shadow() and add_label_missings() gain arguments so you can only label according to the missingness / shadowy-ness of given variables.
• new function which_are_shadow(), to tell you which values are shadows.
• new function long_shadow(), which converts data in shadow/nabular form into a long format suitable for plotting. Related to #165
• Added tests for miss_scan_count

## Minor Changes

• gg_miss_upset gets a better default presentation by ordering by the largest intersections, and also an improved error message when data with only 1 or no variables have missing values.
• shadow_shift gains a more informative error message when it doesn’t know the class.
• Changed common_na_string to include escape characters for “?”, “", "." so that if they are used in replacement or searching functions they don’t return the wildcard results from the characters "?", "”, and “.”.
• miss_case_table and miss_var_table now has final column names pct_vars, and pct_cases instead of pct_miss - fixes #178.

## Breaking Changes

• Deprecated old names of the scalar missingness summaries, in favour of a more consistent syntax #171. The old the and new are:
old_names new_names
miss_case_pct pct_miss_case
miss_case_prop prop_miss_case
miss_var_pct pct_miss_var
miss_var_prop prop_miss_var
complete_case_pct pct_complete_case
complete_case_prop prop_complete_case
complete_var_pct pct_complete_var
complete_var_prop prop_complete_var

These old names will be made defunct in 0.5.0, and removed completely in 0.6.0.

• impute_below has changed to be an alias of shadow_shift - that is it operates on a single vector. impute_below_all operates on all columns in a dataframe (as specified in #159)

## Bug fix

• Ensured that miss_scan_count actually return’d something.
• gg_miss_var(airquality) now prints the ggplot - a typo meant that this did not print the plot

# naniar 0.3.1 (2018/06/10) “Strawberry’s Adventure” 2018-06-08

## Minor Change

This is a patch release that removes tidyselect` from the package Imports, as it is unnecessary. Fixes #174