An identifiable model of two-stage learning

Abstract

An eleven-parameter model for two-stage learning is developed. The model's principal advantage over extant two-stage models is that its parameter space is completely identifiable, thereby eliminating the tedious procedure of locating acceptable identifying restrictions. Identifiability is achieved by defining the model over a slightly modified outcome space. Following the identifiability proof, the necessary statistical machinery for parameter estimation, goodness-of-fit analyses, and hypothesis testing is presented. These latter developments are illustrated with data from an adult cued recall experiment and a free recall experiment with elementary school children.