Latent class models to infer market structure: A comparative analysis

Grover and Srinivasan proposed a latent class model to infer market structure from brand switching data. Jain, Bass and Chen show that, despite different assumptions, it is difficult to empirically distinguish the two models for the switching data on the instant coffee market. Both the models seem to fit the data equally well. In this paper, simulation procedures are used to compare the two models to see how well they recover the true underlying market structure and the parameter values and also identify situations, if any, under which one model fits the data better than the other. Further, the robustness of the factor analytic approach for determining the number of segments in a latent class model is also examined. The results indicate that both the models fit the simulated data adequately well and yield the true overall market structure. However, the recovery of the parameter values is different under the two models, making one procedure more preferable than the other under certain conditions. The results also confirm the approriateness of the factor analytic approach for determining the number of segments in a latent class model.