A study of the competitiveness of eight different estimation algorithms for multinomial logit mode choice modelling using analytical derivatives

The motivation of this thesis is to gain a clear understanding of underlying numerical operational procedures and to identify the differences between eight estimation algorithms adopted for the estimation of the parameters of the Discrete Mode Choice Model of Multinomial Logit (MNL) type. This model is estimated by using the Maximum Likelihood Estimator (MLE). A numerical example is designed to compare the estimation performance of eight algorithms. The data set collected for the study of mode choice decision by airport passengers travelling to airport is used. Due to the limitations of statistical packages, the eight estimation algorithms were coded by using Visual Basic Application (VBA) provided in the Microsoft EXCEL. Results show that the Newton Raphson algorithm is the best of all eight algorithms and dominates its competitors in terms of performance measures. The worst algorithm is the Steepest Ascent algorithm. Also, important factors of econometric model estimation are identified. These are convergence criterion, initial guessing of starting points, and initial Hessian matrix. It is contended that the findings of this research will be very useful to a researcher who has an interest in making a specific code for his/her own model.