Gender identification of water pipits ( Anthus spinoletta) using mixtures of distributions

Abstract

Mixtures of distributions arise when a population is subdivided into homogeneous classes (or components), but the investigator does not know from which of the components any given specimen originates. Mixture analysis is a modern statistical technique to deal with this situation. In particular, mixture analysis attempts to estimate the statistical parameters of each of the components as well as their proportion in the population. This paper reviews some basic concepts of mixture analysis on a non-technical level, and illustrates the method using an example of wing length of water pipits ( Anthus spinoletta ), the two components being males and females. Frequency plots of wing length show distinct bimodality which can be ascribed to sexual dimorphism, but the gender of each individual bird is unknown. Mixtures of two normal distributions appear to be appropriate for modeling this situation.