This research proposes a comparative analysis of the performance of various random utility models (RUM) — namely Multinomial Logit, Nested Logit, Cross Nested Logit, FinMix and CoNL-estimated on a synthetic dataset with variable sample size and correlation patterns. This experimental framework allows comparing model estimates in a fair, controlled environment wherein all relevant characteristics (coefficients, attributes, covariances, likelihood, elasticities) of the “true” underlying model are known. Models are validated especially by comparing true and estimated market share elasticities where the market share is the sum over all observations of the individual probabilities of a given alternative. Indeed, this indicator represents the real forecasting capability of a model, that is the main target for the analyst. Moreover, its true value can be computed when dealing with a synthetic database by evaluating the difference in the number of choices of a given alternative between future and current scenarios, due to a difference in some attribute's value. Comparisons are carried out on several choice contexts, characterized by different correlation matrices and variable sample size.
The importance of choosing appropriate random utility models in complex choice contexts
2017-06-01
350276 byte
Conference paper
Electronic Resource
English
Random and fuzzy utility models for road route choice
Elsevier | 2011
|Random and fuzzy utility models for road route choice
Online Contents | 2011
|Fuzzy vs. random utility models: an application to mode choice behaviour analysis
British Library Conference Proceedings | 2003
|From random utility to random expected utility: theory and application to departure time choice
British Library Conference Proceedings | 2006
|Choosing Public TransportIncorporating Richer Behavioural Elements in Modal Choice Models
Online Contents | 2013
|