DECOMPOSING UNOBSERVED CHOICE VARIABILITY IN THE PRESENCE OF CONSUMERS' TASTE HETEROGENEITY

Heterogeneous tastes across consumers can be captured by random coefficients in a mixed logit (ML) model. However, other types of factors that may not directly affect taste could cause choices to vary, such as choice context, choice task complexity, and demographic characters. This paper jointly considers taste heterogeneity around reference-dependent attributes and other choice variability through inclusion of a scale function, based on data from a stated preference experiment for bread. Results demonstrate that modeling other sources of choice variability in addition to taste heterogeneity increases the model fit, although the improvement is not dramatic.