Measurement of Cannibalism Effects in buying experiments using Mixed Logit Models - The Example of a new Brand of the “Fruits of Lake Constance” Association -

One fundamental assumption of discrete choice regression is the assumption of independence of irrelevant alternatives (IIA). According to the IIA assumption no correlation is allowed between brands in buying experiments. As a consequence, in market simulations all remaining brands gain at the ratio of their starting market share if one brand is excluded from the simulation set. This often does not reflect the reality at the point-of-sale. Mixed-logit models offer the advantage that the IIA-assumption is completely relaxed. What is more, simulations based on mixed logit are able to model cannibalism effects. This paper applies mixed logit to buying behaviour research. A case study is presented where the introduction of a new apple brand at the German discounter “Penny” is simulated in a buying experiment.