Statistical models as data structures.
23 October 1986
This paper describes the early stages of research aiming at a unified approach to computations for statistical models, particularly for a variety of nonlinear models. The computations include fitting models to data, examining the models numerically or graphically, symbolic computations, and inferential methods (such as resampling). We unify the computations by defining models as a class of data structures, with attributes providing the parameters, statistical data and formal definition of a particular model. Our approach is closely related to ideas of "data-driven" or "object-oriented" programming. We illustrate the approach through brief discussions of fitting and resampling models, and of model evolution.