Example: Missing Values

What a missing value means

A missing value records information that is unavailable. It is different from a measured zero. R represents missing values with NA, without quotation marks; "NA" is ordinary text. Although the constant NA is logical, R converts it to the appropriate type when it is placed in a character or numeric vector.

x <- c("A", "B", NA, "D")
is.na(x)

[1] FALSE FALSE  TRUE FALSE

is.na() asks whether each entry is missing. Use it instead of x == NA: a comparison with an unknown value does not establish equality.

Count before making a decision

The following invented table contains three measurements. Create it directly in R; no download is required.

measurements <- data.frame(
  ID = c("01", "02", "03"),
  Height_m = c(18.5, NA, 16.2),
  stringsAsFactors = FALSE
)
colSums(is.na(measurements))

      ID Height_m
       0        1

is.na(measurements) produces a logical table and colSums() counts the missing entries in each column. Inspect the affected rows before deciding what to do:

measurements[is.na(measurements$Height_m), ]
mean(measurements$Height_m)
mean(measurements$Height_m, na.rm = TRUE)

The first mean is NA. With na.rm = TRUE, the mean is 17.35 m, calculated from the two available heights. Report that one of three measurements was unavailable: excluding it does not recover its value or remove possible bias.

Keep numerical columns numerical. Replacing an unavailable height with "no value" would turn the entire column into text. A display label can be created in a separate character column if needed.

Preserve source symbols

Official tables can use symbols such as -, ., or /. Their meanings depend on the source documentation. Inspect them as text before converting a column to numbers. Preserve the original tokens so that different reasons for an unavailable number remain distinguishable.

reported <- c("18,5", ".", "16,2")
table(reported, useNA = "ifany")

If the documentation says that . represents an unavailable number, create a numerical version while retaining reported:

height <- reported
height[height == "."] <- NA
height <- as.numeric(sub(",", ".", height, fixed = TRUE))

This example explains a conversion rule; it does not assign meanings to every symbol in the regional crop file. An unexplained token should lead to a documented question, not an automatic replacement with zero.

Apply this to the regional data

In the worked cleaning example and crop assignment:

  • count source symbols and missing entries before and after conversion;
  • explain the chosen treatment for each documented symbol;
  • retain source codes alongside numerical data where they carry additional information;
  • distinguish an optional missing descriptor from an unavailable measurement;
  • explain exclusions and check how many observations remain for analysis.

A table with no missing cells can still be wrong. The goal is a dataset whose values and limitations you can explain. See the R documentation for missing values.

Updated: