A comparison of approaches to exploit budget allocation data in cross-sectional maximum likelihood estimation of multi-attribute choice models

In this paper, four calibration approaches to exploit budget allocation data in maximum likelihood estimation of multi-attribute choice models are proposed. They differ on the implicit meaning of the dependent variable: (A) share of consumers according to the preferred alternative; (B) share of sales; (C) average share of consumer's budget; and (D) share of sales according to the preferred alternative. Differences between them can be conceived as depending on two circumstances: customer loyalty and customer selectivity. These are tested in the context of spatial consumer behavior, market response to hypermarket chains being represented as a function of their location strategies. Results show that different nuances on the definition of the dependent variable lead to slightly different relationships with the explanatory variable and to different predictive capabilities.