Marcela Munizaga

Hybrid Choice Models: Progress and Challenges

We discuss the development of predictive choice models that go beyond the random utility model in its narrowest formulation. Such approaches incorporate several elements of cognitive process that have been identified as important to the choice process, including strong dependence on history and context, perception formation, and latent constraints. A flexible and practical hybrid choice model is presented that integrates many types of discrete choice modeling methods, draws on different types of data, and allows for flexible disturbances and explicit modeling of latent psychological explanatory variables, heterogeneity, and latent segmentation. Both progress and challenges related to the development of the hybrid choice model are presented.

Representation of heteroskedasticity in discrete choice models

The Multinomial Logit, discrete choice model of transport demand, has several restrictions when compared with the more general Multinomial Probit model. The most famous of these are that unobservable components of utilities should be mutually independent and homoskedastic. Correlation can be accommodated to a certain extent by the Hierarchical Logit model, but the problem of heteroskedasticity has received less attention in the literature. We investigate the consequences of disregarding heteroskedasticity, and make some comparisons between models that can and those that cannot represent it. These comparisons, which use synthetic data with known characteristics, are made in terms of parameter recovery and estimates of response to policy changes. The Multinomial Logit, Hierarchical Logit, Single Element Nested Logit, Heteroskedastic Extreme Value Logit and Multinomial Probit models are tested using data that are consistent with various error structures; only the last three can represent heteroskedasticity explicitly. Two different kinds of heteroskedasticity are analysed: between options and between observations. The results show that in the first case, neither the Multinomial Logit nor the Single Element Nested Logit models can be used to estimate the response to policy changes accurately, but the Hierarchical Logit model performs surprisingly well. By contrast, in a certain case of discrete heteroskedasticity between observations, the simulation results show that in terms of response to policy variations the Multinomial Logit model performs as well as the theoretically correct Single Element Nested Logit and Multinomial Probit models. Furthermore, the Multinomial Logit Model recovered all parameters of the utility function accurately in this case. We conclude that the simpler members of the Logit family appear to be fairly robust with respect to some homoskedasticity violations, but that use of the more resource-intensive Multinomial Probit model is justified for handling the case of heteroskedasticity between options.

Detection of Activities of Public Transport Users by Analyzing Smart Card Data

During the past decade, a significant amount of research has been dedicated to the use of smart card data for various purposes. A method is presented for the detection and estimation of the location, time, duration, and purpose of activities undertaken by public transit users with the use of smart card databases and other available information about land use and user behavior. The method is applied to cases in Santiago, Chile, and Gatineau, Quebec, Canada, to identify activity purpose and time frame to characterize user behavior. The results obtained for each city are compared to discover differences in behavioral activity patterns due to sociological, cultural, and geopolitical differences.

Testing mixed logit and probit models by simulation

Discrete choice models with error structures that are not independent and identically distributed have received enormous attention in the recent literature. A detailed synthetic study tests this type of model in a controlled case. With mixed logit and probit models as the study objects, calibration was implemented with the use of software available on the Internet. The controlled situation was built as a simulation laboratory, which generated databases with known parameters. The effects of various elements were analyzed: number of repetitions of the simulation, number of observations in the database, and how the use of Halton sequences improves the mixed logit calibration. The scale effects on the different models are also discussed. The results obtained in this specific context lead to some recommendations for future users of these powerful modeling tools. In particular, flexible structures require large sample sizes to calibrate the elements of the covariance matrix.

Valuing Time With a Joint Mode Choice-Activity Model

Joint estimation is called for in the common framework shared by activity and travel models. In literature, attention has only recently been given to this complex problem. Various components of times subjective value (value of assigning time to travel, resource value, and travel time savings) have been allowed to be disentangled after obtaining travel choice and activity duration explicit equations. The only reported results have been preliminary results assuming independence, although expectation of some interrelation between equations is natural. The authors postulate a general error structure and adapt a discrete/continuous econometric model. The authors obtain more robust models, and significantly different and more credible value of time estimates.

Econometric Calibration of the Joint Time Assignment--Mode Choice Model

This paper describes the derivation and the econometric calibration of a joint time assignment--mode choice model with a microeconomic foundation, to be applied to the TASTI (time assignment travel and income) database. The econometric procedure is a full information maximum likelihood with three nonlinear continuous equations and one discrete choice. We use Lees transformation [Lee, L. F. 1983. Generalized econometric models with selectivity. Econometrica51 507--512] to include correlations between the continuous and discrete equations. This allows us to estimate (a) the value of time as a resource or value of assigning time to a pleasurable activity, (b) the value of assigning time to work, and (c) the value of assigning time to travel. We apply the method and obtain reasonable results. Finally, we identify some econometric challenges for further research.

Indirect Measurement of Level of Service Variables for the Public Transport System of Santiago Using Passive Data

Abstract Purpose — The introduction of new technology to public transport systems has provided an excellent opportunity for passive data collection. In this paper, we explore the possibility of automatically generating level of service indicators that could be used for operation planning and monitoring of Transantiago, the public transport system of Santiago, Chile. Design/methodology/approach — After basic processing of the raw automatic vehicle location (AVL) and automatic fare collection (AFC) data, we were able to generate bus speed indicators, travel time measurements and waiting time estimates using data from 1week. The results were compared with manual measures when available. Findings — The advantage is that these measurements and estimates are reliable because they are obtained from large samples and at nearly no cost. Moreover, they can be applied to any set of data with a selected periodicity. Research limitations — The scope of this research is limited to what can be observed with AVL and AFC data. Additional information is required to incorporate other dimensions, such as personal characteristics and/or more detail in the origin/destination (OD) of the trips. Practical implications — Nevertheless, these results are valuable for the planning and operation management of public transport systems because they provide large amounts of information that is difficult and expensive to obtain from direct measurements. Originality/value — This paper proposes tools to obtain valuable information at a low cost. These tools can be implemented in many cities that have certain technological devices incorporated into their public transport systems.

Measuring User Satisfaction in Transport Services: Methodology and Application

Abstract Purpose — New methods of measuring user satisfaction in transport services have been proposed and applied in the literature. In this paper, we compare three alternative measures for estimating user satisfaction: the numerical rating, the ordinal rating and the choice. Approach — We analysed these measures considering their differences and limitations and the models that use these measures as dependent variables. We developed and applied a methodology to build these models. It comprises a preliminary qualitative analysis and a quantitative survey to identify the most relevant attributes of the satisfaction function, and a stated preference survey to obtain information of the alternative satisfaction measures for modelling purpose. Findings — The ordinal rating may be a better user response to estimate satisfaction than score and choice based on its characteristics. The results obtained in the application reinforced this approach. Research limitations — It is assumed that choice, score and ordinal valuation depend upon a latent stochastic satisfaction function of the same attributes. Further research is needed to analyse this assumption and how these responses vary according to the context for decision and exogenous factors, including the response scale of ratings. Practical implications — Gathering alternative satisfaction responses simultaneously from users allowed for the consistency analysis and filtering of data, which greatly benefited the model estimation process. Originality/value — The paper provides a methodology to estimate user satisfaction models in transit services, which can be applied in other transport services. The conceptual analysis and the application suggest that ordinal ratings are key user responses to uncover the underlying satisfaction function.

Towards a Reliable Origin-Destination Matrix from Massive Amounts of Smart Card and GPS Data: Application to Santiago

Abstract Purpose — Automated fare collection systems implemented in public transportation systems in the last decade have provided a massive, continuous and low-cost source of reliable travel information. A direct and useful application of these data is the estimation of highly representative, although not bias-free, origin-destination (OD) matrices. Methodology/approach — We discuss several issues with current OD matrix estimation methodologies, such as fare evasion and group travel, and their derived biases, specifically focusing on the Santiago (Chile) case. We also propose and apply two methods of validation: endogenous and exogenous validation. We elaborate on some methodological improvements that could be implemented to upgrade the activity estimation mechanics. Findings — Several sources of bias in the estimation of OD matrix estimation from passive data are pointed and some solutions proposed. We apply improvements to existing methodologies and increase the success rate of trip estimations. Practical implications — The reliable estimation of public transport OD matrices from passive data results in a valuable planning tool for both transit authorities and operators, much more representative and with less errors and biases that conventional data collecting techniques. Originality/value of paper — This paper is one of the first works to deal with the subject.

Bus Arrival Time Prediction with Limited Data Set using Regression Models.

The increase of population has intensified everyday rush. Traffic congestions are still a problem in cities and are one of the main cause for public transport delays. City residents and visitors have experienced time loss by using public transport buses, because of waiting at the bus stops and not knowing if the bus is delayed or already serviced the stop. Therefore it is valuable for people to know at what time the bus should arrive (or is it already missed) at specific bus stop. Real-time public bus tracking and management system development has been the focus of many researchers, and many studies have been done in this area. This paper focuses on bus travel time prediction comparison between linear regression and support vector regression models (SVR), when using limited data set. Data were limited in a way that only historical GPS (Global Positioning System) coordinates of bus location (recorded each 30 seconds) and driven distance were used, there were no information about arrival/departure times, delays or dwell times. Distance between stops and delay (assumed values based on route observations by authors) were used as inputs for both models. It was concluded that SVR algorithm showed better results, but the difference was not significantly large.