Speakers: Thomas Magor and Len Coote
Affiliation: UQ School of Business


In 2000, Daniel McFadden – Professor Emeritus at UC Berkeley – was awarded the Nobel Prize in Economic Sciences for theory and analysis of discrete choice.  McFadden’s model – conditional (multinomial) logit – retrieves the aggregate preferences of decisions makers.  Extensions of his model – the random coefficient and error component forms of mixed logit – in addition retrieve unobserved (latent) sources of preference heterogeneity.  These models are estimated using maximum likelihood and maximum simulated likelihood approaches, respectively. 

McFadden’s model and its extensions are used in applied economics, including environmental, health, and transport economics (and in quantitative marketing) – principally for forecasting demand for discrete choices.  McFadden’s model and extensions of his model are fit to the microdata of individual choices.  The microdata may be marketplace choices (“revealed preferences” – preferences revealed by marketplace choices) or choices among hypothetical alternatives in an experiment (stated preferences). 

Our contribution – in a series of papers in the Journal of Choice Modelling – is the development of a very general and flexible form of mixed logit (or “structural choice model”).  Structural choice modelling (SCM) is specifically designed to incorporate latent variables (representing unobserved sources of preference heterogeneity) and subsumes many existing choice models as special cases.  Applications of SCM include the study of multiple data generation processes: dynamics, embedded experiments, complex decision making, and group decision making. 

We think there are many potential collaboration opportunities with researchers in statistics.  Some immediate opportunities may include: (1) exploring gradient-based methods for structural choice model estimation other than the standard Newton-Raphson method; (2) studying the performance of different estimation approaches for SCM (e.g., maximum simulated likelihood versus composite marginal likelihood); and (3) developing new estimation software for SCM (i.e., creating an R package that can be disseminated to researchers in applied economics). 

About Statistics, modelling and operations research seminars

Students, staff and visitors to UQ are welcome to attend our regular seminars.

The events are jointly run by our Operations research and Statistics and probability research groups.

The Statistics, modelling and operations research (SMOR) Seminar series seeks to celebrate and disseminate research and developments across the broad spectrum of quantitative sciences. The SMOR series provides a platform for communication of both theoretical and practical developments, as well as interdisciplinary topics relating to applied mathematics and statistics.


Chamberlain Building (35)
Room: 102 and via Zoom (https://uqz.zoom.us/j/85172010876)