Aggregation Bias in Discrete Choice Models with an Application to Household Vehicle Choice

This paper studies the practice of aggregating choices within discrete choice models. Researchers often do not observe choices at the exact level they are made, and hence aggregate choices to the level that is observed. Modeling choices at a fine level of detail can also lead to large choice sets that exceed the practical capabilities for model estimation. However, the practice of aggregation misspecifies the true choice set of interest. We investigate this concern within the context of the Berry, Levinsohn, and Pakes (BLP) choice model for microand macro-level data. We compare the practice of aggregating choices to specifications from two papers that address these concerns (McFadden, 1974; Brownstone and Li, 2014), with application to vehicle choice data. We find that aggregation affects both the point estimates and standard errors obtained from the model. In particular, standard errors are smaller with aggregation. This result has significant empirical implications. Discrete choice models are widely used to estimate consumer valuation of fuel efficiency, a quantity that is relevant to energy analysts concerned that consumers undervalue fuel efficiency technologies (the “energy paradox”). If so, then there is space for policies that increase adoption of such technologies. However, estimates of consumer valuation across vehicle choice studies are inconclusive. The findings of this paper suggest that this disparity may be partly explained by the practice of aggregating choices. In addition, the BLP model applied here is usually estimated sequentially, and the standard errors derived from this process are inconsistent. Thus, this paper also derives consistent standard errors for the model and examines their performance compared to the sequential standard errors that are commonly used.