Extreme Value Analyses with Missing Data

Performs likelihood-based extreme value inferences with adjustment for the presence of missing values. A GEV distribution is fitted to block maxima using maximum likelihood estimation, with the GEV location and scale parameter reflecting the numbers of non-missing raw values in each block. A Bayesian version is also provided. For the purposes of comparison, there are options to make no adjustment for missing values or to discard any block maximum for which greater than a percentage of the underlying raw values are missing. A plot method provides a set of standard model diagnostic plots, with appropriate adjustment made for the presence of missing values. Example datasets containing missing values are provided.

The evmissing package was created to accompany Simpson and Northrop (2026).

An example

The main function in evmissing is gev_mle(), which fits a GEV distribution to block maxima using maximum likelihood estimation, with the option to make an adjustment for the numbers of non-missing raw values in each block.

Our adjustment is based on the strong assumption that missing values occur completely at random. We suppose that a block maximum