Using mixtures of Weibull distributions to estimate mixing proportions (Q1822160)

Mixture distributions characterize many physical measurements in which observable variates are generated from one of several component distributions. It is common to model such measurements as mixtures of normal distributions and often the primary interest is in the estimation of the mixing proportion(s). It is demonstrated that asymmetry in the component distributions can cause a substantial bias in these normal-based proportion estimators. We examine the use of minimum distance estimation based on component distributions belonging to the Weibull family, whose members can be either symmetric or skewed depending on parameter configurations. Minimum distance estimation in this setting provides a robust and computationally efficient alternative to the normal-based techniques currently in use.