Median with Missing Values Analysis

The goal of naidem is to compute the median in a way that handles missing values properly: returning NA if and only if the median can’t be determined from the data. Its source code has no dependencies.

Use median2() as a drop-in replacement for median(). It is the same as in median() by default, except for handling missing values.

Installation

You can install the development version of naidem like so:

remotes::install_github("lhdjung/naidem")

Get started

library(naidem)

Why and how to use naidem:

The problem

Base R’s median() function returns NA whenever the input vector contains one or more NAs. In many cases, missing values do make it impossible to compute the median. Yet some distributions have a clear median even so:

x1 <- c(6, 7, 7, 7, NA)
median(x1)
#> [1] NA

The actual median is 7, irrespective of the true value behind NA.

The solution

Use naidem’s median2() instead. This function will return the median whenever it can be determined:

median2(x1)
#> [1] 7

If the median really is unclear, both functions return NA:

x2 <- c(3, 4, 4, 5, NA, NA)
median(x2)
#> [1] NA
median2(x2)
#> [1] NA

Compare this to NA^0, which returns 1 even though NA^2 returns NA. If the exponent is 0, the result is the same for all possible bases. The same is sometimes true for the median. Therefore, implementations should check for this case and return the median if and only if it can be determined. This also makes NA more meaningful when it is returned: users can be sure that the median really is unknown.

See Implementing the algorithm for more information on naidem’s solution.

Possible medians

What to do if the median really is unknown, like above? median_range() will attempt to narrow it down to its minimal and maximal possible medians:

median_range(x2)
#> [1] 3.5 4.5

About this package