Some Comments on the Development and Application of Linear Learning Models

This paper evaluates some learning models and explores the possibility of extending the linear learning models by adding explanatory variables, using Lilien's model as a point of departure. Two parameter estimation schemes are proposed and applied: one at the micro-level using consumer panel data (micro-data) and the other at the macro-level using aggregate time-series data (macro-data). The use of micro-data restrains the estimation of effects to a limited number of decision variables in Lilien's model, because of a lack of degrees of freedom. It is shown that by modifying the equations of Lilien's model and by considering more than two brands the number of degrees of freedom can be increased considerably. This opens the possibility of estimating the response parameters of many more explanatory variables, in theory. However, practically speaking, estimation remains difficult because the parameters in the modified model are heavily restricted by standard probabilistic and other constraints. The endeavor to estimate the parameters of the modified linear learning model using macro-data is not feasible. This leads to the conclusion that the application of learning models in the area of decision-making in marketing is rather difficult and limited.