Skip to contents

Missing Values Policy


Compositional data are quantitative positive descriptions of the parts of some whole, carrying relative, rather than absolute, information (ie. only relative changes are relevant; Aitchison 1986).

Basically, three situations can be outlined regarding missing values in compositions:

  • Unobserved quantities.

  • Amounts observed, but which happen to be below the detection limit (thus interpreted as small unknown values).

  • Absolutely zero quantities.

These situations can be represented in several ways:

  • The presence of zeros.

  • The presence of missing values (NA).

When creating a CompositionMatrix object, the presence of zero and NA values is allowed: this makes it possible to explore and visualize the data while preserving the missing structure. However, the user must deal with these missing values before proceeding further (e.g. by removing incomplete cases or replacing the values concerned): log-ratio transformations cannot be computed in the presence of zeros or missing values.


If you need more advanced features (e.g. imputation of missing values), you should consider the compositions or robCompositions package.


Aitchison, J. (1986). The Statistical Analysis of Compositional Data. London: Chapman and Hall.

See also

Other imputation methods: replace_NA(), replace_zero()