Modeling Heterogeneity in Choice Models, Household Level vs. Intra-household Heterogeneity in Reference Price Effects: Should National Brands Care?

Much of the extant empirical work on consumers’ grocery purchases employ models that are estimated on household scanner panel data. A known limitation of these models is that households may have multiple decision makers, and a decision maker may have brand preferences and marketing mix sensitivities that are distinct from other decision makers in the household. We seek to study whether models using individual customer data provide substantially different insights and managerial implications relative to models that use household data. This important issue has not been addressed in the literature, possibly due to limitations of scanner panel data. Using a unique data set that identifies choices made by individual customers within a household, we estimate multinomial choice models at the household level with and without incorporating intra-household heterogeneity using Markov Chain Monte Carlo (MCMC) procedures. We incorporate controls for unobserved heterogeneity by estimating random coefficients models which allows the brand preferences and the price sensitivity parameters to vary across households. We find that in each product category the estimates obtained at the customer level are significantly different from those obtained at the household level. Our findings imply that targeting promotions based on customer level estimates will result in outcomes that are significantly more profitable relative to targeting based on household level estimates.