Missing values

In real data, something is always missing: someone skipped a question, a measuring device failed, a question was never asked. R writes NA for this, for not available. How R deals with it is annoying at first sight and very sensible at second.

NA is contagious

With an unknown value, every result is unknown too. What is 5 plus a number nobody knows? R can only answer that with NA:

organic_share <- c(5, NA, 2, 4)
mean(organic_share)

This also holds for comparisons. Even NA == NA gives NA, because nobody knows whether two unknown values are equal.

Leaving missing values out

Many functions have the argument na.rm (remove NA). With na.rm = TRUE they only calculate with the values that are there:

organic_share <- c(5, NA, 2, 4)
mean(organic_share, na.rm = TRUE)

Having to say so explicitly is deliberate: missing values should not vanish unnoticed. The mean above now refers to three people, not four.

Finding missing values

Because a comparison with NA does not work, there is a function of its own for this. is.na() gives TRUE where a value is missing:

organic_share <- c(5, NA, 2, 4)
is.na(organic_share)

Together with what you know about counting, sum(is.na(organic_share)) gives the number of missing values.

Ask first why something is missing

Before you leave missing values out, it is worth asking why they are missing. In a survey, NA often means "not asked", because a question was put to only some of the respondents. If you simply remove such values, you are afterwards talking about a different group of people than before.

Key points

• NA is an unknown value; every calculation with it gives NA again.
• na.rm = TRUE explicitly leaves missing values out.
• is.na() finds them, sum(is.na(x)) counts them.