Bimodality is observed in empirical distributions of variables related to materials (glass resistance), to companies (productivity) and to natural phenomena (geyser eruptions). Modeling bimodality is usually done by means of mixtures of statistical distributions. Mixtures represent a suitable model when two or more distinct groups with specific distributions are considered as a single set. Conversely, when original groups composing the mixture are not recognizable, or when the observed phenomena is structurally bimodal, it is worth using an alternative approach to obtain a bimodal distribution fitting the data.

Our proposal exploits the change of variable theorem which requires: 1) the choice of a generating density function representing the main features of the phenomena under analysis; 2) the choice of a transforming function that describes the observed departure from the expected behavior. Such procedure has been frequently used since '50s only for normalizing unimodal phenomena.

The novelty of this work consists in putting particular attention to the choice of the transformation in two different cases: when bimodality arises from a slight departure from unimodality and when it is a proper structural feature of the variable under study. As an example we use the R geyser dataset.