Parameter Estimation Methods for Discrete Exponential Learning Curves

A discrete exponential model has been proposed for use as a learning curve. This model describes different learning behavior than the classical or a modified version of the power-form model and it provides some practical advantages. Tests for selection between the models are given. Methods of parameter estimation for the exponential model based on least-squared errors, maximum likelihood and mean likelihood criteria are shown for individual cycle time data. Estimation methods are also given for data on the number of completed cycles during a fixed time interval. Monte Carlo simulation experiments were conducted to test the past description and future forecast quality of this model using these estimation methods. The classical power-form model was also included in these experiments. Estimation precision is described for both models. Other features of the exponential model are shown in contrast to the power-form models.