Finite mixture coding: A new approach to coding continuous characters

Abstract

Finite mixture coding (FMC) is a new method of coding continuous characters. FMC uses a three-step goodness-of-fit procedure to assign codes. First, for a given measurement, parameters are estimated for a number of density functions that describe a data set either of species means or of measurements of specimens from several species. The density functions represent either a single population or a mixture of populations (e.g., a mixture of two normal distributions). Next, a goodness-of-fit criterion (the Akaike information criterion) is used to determine which of the density functions best describes the data set. The best function indicates the number of populations into which the variates of the data set can be segregated. Finally, species are assigned to the population for which its probability of membership is highest. Each population is then assigned a code, and species falling within the same population share the same code. Although other coding methods incorporate statistical tests or parameters into the coding process, FMC is the only method that produces codes as the direct output of a statistical procedure.