Is sequential estimation a suitable second best for estimation of hybrid choice models?

Abstract

The simultaneous estimation method has overtaken the sequential approach as the preferred estimation method for hybrid discrete choice models. Notwithstanding, the computational cost of the simultaneous estimation can be prohibitive when models become more involved, and in such cases sequential estimation can still be a potent option. In previous work a theoretical analysis was conducted that led to the identification of a major bias affecting the sequential estimation method, and a correction term was proposed for the bias induced on the estimated parameters by the variability associated with the latent variables. However, no attempt was made to quantify that induced variability. In this study, an attempt was made to determine the nature of the variability induced through the latent variables as well as the viability of relying on the sequential estimation method as an alternative (second-best) estimation tool for cases in which the complexity of the specification makes relying on simultaneous estimation unfeasible. Results show that the sequential method behaves in an acceptable way (the bias can be avoided through the correction) when the variability associated with the latent variables is low in comparison with the error term of the discrete choice model. However, when this variability is considerable the bias correction becomes an intricate matter and appropriate results cannot be guaranteed.