A New Model for Learning Curves, DARM

This article introduces a new model for learning curves, a displaced autoregressive model (DARM). The model has theoretical and practical advantages over both the traditional exponential learning-curve model and the exponential leaming-curve model augmented by an autoregressive disturbance process. The DARM parameters can be related to the characteristics of the learning process and are easy to estimate. DARM appears to fit the data better than the alternative models, reasonable estimates of long-run average item costs can be obtained from a few observations, and the model can be modified to allow for learning plateaus.