Assessing User Benefits with Discrete Choice Models: Implications of Specification Errors Under Random Taste Heterogeneity