The recoverability of segmentation structure from store-level aggregate data

The authors focus on estimating a latent-class choice model with consumer response segments when only store-level aggregate data are available. Most of the proposed methodologies in the marketing literature require household panel data, which can be difficult to obtain. There is a growing stream of work in marketing and in empirical industrial organization that estimates segmentation structure with aggregate data. This article is a careful attempt to understand the extent to which disaggregate structure in the form of a latent-class model can be recovered from aggregate data. The authors show that under specific assumptions and when the household-level model is correctly specified, most of a latent-class segmentation structure is identifiable even if only store-level aggregate data are available. Therefore, the store data-based estimates for the latent-class model are consistent. In other words, the mean absolute deviation (MAD) of the estimates goes to zero with infinite sample size. To assess how well latent-class structure can be estimated from store data sets of sample sizes that are comparable to those in real life, the authors simulate more than 60,000 store data sets and compute the consequent model estimates and their estimation errors. The results show that the MAD of the latent-class estimates diminishes much more slowly with store data than with household data. Moreover, the rate is so slow that obtaining estimates with reasonably small MADs often requires unreasonably large sample sizes. The authors' simulations offer guidance on conditions that favor obtaining more accurate estimates from store-level data.