MODELING TASK COMPLETION DATA WITH INVERSE GAUSSIAN MIXTURES

Whitmore introduced mixtures of inverse Gaussian distributions and fitted them to several data sets involving duration phenomena. In this paper we model task completion data from a large automobile plant in southern Ontario, with a mixed inverse Gaussian distribution. The mixing is over a parameter that can be naturally identified with the intensity with which the work crew approaches its task. It is found that, depending on the nature of the task, the mixed distribution may provide an improvement over the familiar pure inverse Gaussian distribution. A secondary consideration is the study of asymptotic likelihood inference for such mixtures.