Relative efficiency appraisal algorithms using small-scale through empirically tailored of discrete choice modeling maximum likelihood estimator computing environment

摘要

Discrete choice models are widely used in multiple sectors such as transportation, health, energy, and marketing, etc., where the model estimation is usually carried out by using commercial software. Nonetheless tailored computer codes offer modellers greater flexibility and control of unique modelling situation. Aligned with empirically tailored computing environment, this research discusses the relative performance of six different algorithms of a discrete choice model using three key performance measures: convergence time, number of iterations and iteration time. The computer codes are developed by using Visual Basic Application(VBA). Maximum likelihood function(MLF) is formulated and the mathematica relationships of gradient and Hessian matrix are analytically derived to carry out the estimation process. The estimated parameter values clearly suggest that convergence criterion and initial guessing of parameters are the two critical factors in determining the overall estimation performance of a custom-built discrete choice model.